{"schema":"5lc-index-ledger/1","key":"chips-chip-design","layer":"chips","sublayer":"chip-design","asOf":"2026-10-22","about":"5LC Importance Index ledger for one sublayer: the measure card, the current ten with any editorial overrides and their reasons, the computed result for every scored company (Importance = share of AI output × years to replace, Reach, replacement years, Rising, the Monte Carlo rank simulation and its stability class, validation, the revenue baseline and rater agreement), each scored company's sourced inputs and irreplaceability ratings, and the sublayer's validation cases with their sources. Monte Carlo frequencies describe the stated input ranges, not true probabilities. Omitted: the free-text notes of the internal check passes and of rating reconciliations, and the location of the internal card review; references to local evidence files and to internal method decisions are replaced by plain descriptions.","roster":{"members":[{"entityId":"nvidia","slot":"importance","since":"2026-10-22"},{"entityId":"broadcom","slot":"importance","since":"2026-10-22"},{"entityId":"huawei","slot":"importance","since":"2026-10-22"},{"entityId":"amd","slot":"importance","since":"2026-10-22"},{"entityId":"alphabet","slot":"importance","since":"2026-10-22"},{"entityId":"amazon","slot":"importance","since":"2026-10-22"},{"entityId":"marvell","slot":"importance","since":"2026-10-22"},{"entityId":"intel","slot":"importance","since":"2026-10-22"},{"entityId":"cambricon","slot":"rising","since":"2026-10-22"},{"entityId":"credo","slot":"rising","since":"2026-10-22"}],"pending":{},"lastRun":"2026-10-22"},"card":{"schema":"5lc-index-card/1","key":"chips-chip-design","layer":"chips","sublayer":"chip-design","version":1,"scope":"AI data-centre chips at the companies that design them, counting both the owner and the design partner of a custom chip; device chips are out, and chip manufacturing, memory, packaging, networking-equipment makers, design software and non-processor IP belong to their own sublayers.","outputMeasure":"Value of AI data-centre chips supplied in 2025, at designers' revenue or, for custom chips, owners' cost.","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","hardToReplace":"Leading accelerators need years of design, mature software and reserved manufacturing capacity, and custom chips tie owners to one design partner per generation. Processors and networking chips have more alternatives.","functions":[{"id":"merchant-accelerators","name":"Merchant accelerators","why":"Most AI training and serving runs on them.","weight":{"low":0.682,"central":0.711,"high":0.741},"weightSource":{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","title":"NVIDIA Corporation Form 10-K for the fiscal year ended January 25, 2026","publisher":"NVIDIA / U.S. SEC","publishedDate":"2026-02-25","accessedDate":"2026-10-22","tier":"T1","locator":"Notes to the Consolidated Financial Statements, revenue by specialized markets: Data Center Compute $162,361 million in fiscal 2026"},"shareData":"Accelerator revenue from filings, or shipment estimates"},{"id":"custom-accelerators","name":"Custom accelerators","why":"Cloud and AI companies' own chips, made with partners; cheaper per chip.","weight":{"low":0.059,"central":0.067,"high":0.083},"weightSource":{"url":"https://www.fool.com/earnings/call-transcripts/2025/12/12/broadcom-avgo-q4-2025-earnings-call-transcript/","title":"Broadcom (AVGO) Q4 2025 earnings call transcript","publisher":"The Motley Fool (transcript of Broadcom's earnings call)","publishedDate":"2025-12-12","accessedDate":"2026-10-22","tier":"T2","locator":"Hock Tan, prepared remarks: fiscal 2025 AI revenue grew 65% to $20 billion; he said on the Q3 FY2025 call that XPUs were 65% of AI revenue"},"shareData":"Programme value, split equally between the owner and its design partner"},{"id":"networking-silicon","name":"AI networking chips","why":"Join accelerators into clusters.","weight":{"low":0.142,"central":0.181,"high":0.19},"weightSource":{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","title":"NVIDIA Corporation Form 10-K for the fiscal year ended January 25, 2026","publisher":"NVIDIA / U.S. SEC","publishedDate":"2026-02-25","accessedDate":"2026-10-22","tier":"T1","locator":"Notes to the Consolidated Financial Statements, revenue by specialized markets: Data Center Networking $31,376 million in fiscal 2026"},"shareData":"Networking-chip revenue from filings and market-share reports"},{"id":"host-cpus","name":"Host processors","why":"Run and feed every accelerator server.","weight":{"low":0.028,"central":0.039,"high":0.052},"weightSource":{"url":"https://www.trendforce.com/presscenter/news/20250106-12433.html","title":"TrendForce: AI Server Market Will Keep Growing Through 2025, with Related Industry Value Projected to Reach US$298 Billion","publisher":"TrendForce","publishedDate":"2025-01-06","accessedDate":"2026-10-22","tier":"T4","locator":"AI servers' portion of overall server shipments to increase to more than 15% in 2025"},"shareData":"Share of processors in AI servers, including NVIDIA Grace"},{"id":"processor-ip","name":"Processor IP","why":"Licensed designs inside Arm-based and custom chips.","weight":{"low":0.0016,"central":0.002,"high":0.004},"weightSource":{"url":"https://www.sec.gov/Archives/edgar/data/1973239/000197323926000097/arm-20260331.htm","title":"Arm Holdings plc Form 20-F for the fiscal year ended March 31, 2026","publisher":"Arm Holdings / U.S. SEC","publishedDate":"2026-05-26","accessedDate":"2026-10-22","tier":"T1","locator":"Item 5.A Results of Operations: royalty revenue $2,613 million and licence and other revenue $2,307 million in fiscal 2026"},"shareData":"Data-centre royalty and licence revenue"}],"risingFloorPct":0.25,"risingFloorReason":"One supplier holds most merchant-accelerator value, so at 1% only the eight Importance companies would qualify; 0.25% lets fast-growing smaller designers compete for the two Rising places.","validationCaseIds":["chips-chip-design-nvidia-accelerator-allocation-2023"],"knownGaps":["AMD, Huawei and most Chinese and private designers do not report accelerator revenue; the weights use estimates (AMD's data-centre segment less an estimate for server processors; Huawei from reported shipment estimates).","Custom chips have no market price: their value is what owners pay design partners, which leaves out memory or wafers an owner buys directly and the owner's own design work.","NVIDIA sells networking mostly as switches, adapters and cables, while Broadcom, Marvell and others sell chips, so revenue shares in networking mix product levels.","NVIDIA Grace processors are sold inside combined processor-accelerator modules; their value is estimated, moved to host processors and removed from accelerators.","No source reports each vendor's share of processors in AI servers; the AI-server share of processor revenue is estimated from shipment data.","Arm does not report data-centre royalties in dollars; the processor IP total is estimated from its statements.","Fiscal years differ (NVIDIA and Marvell end in January, Broadcom in early November, Credo in late April or early May), so 2025 values are taken from the fiscal year closest to calendar 2025."],"approval":{"approver":"Dachi","date":"2026-10-22","sha256":"36ac5c90b22b681a90c364035c6720d9720e0d054e5b50b44c14deefd00b7646"}},"result":{"schema":"5lc-index-result/1","key":"chips-chip-design","asOf":"2026-10-22","methodVersion":"1.1","methodSha256":"3cabed6b38514c263d8b86a494e8cc19cdbd0b2a1013543dc024192c49f7a6a6","cardSha256":"36ac5c90b22b681a90c364035c6720d9720e0d054e5b50b44c14deefd00b7646","validationSha256":"409931bb458f38b02342c9bf43e61eab299e47d8679912af2c96074ec556f17e","inputsSha256":"777afc472c5dfb0d400f90bfc1c3fe68b00c40aaf59fdfe02ecc9bf81ea2c113","unscreened":[],"companies":[{"entityId":"nvidia","name":"NVIDIA","reachPct":77.8537,"importance":142.964098988,"importanceRank":1,"functions":[{"id":"merchant-accelerators","weight":0.711,"sharePct":90.11,"shareLowPct":89.26,"shareHighPct":90.68,"stale":false,"offBasis":false,"level":4,"years":2,"disagreement":false},{"id":"custom-accelerators","weight":0.067,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"networking-silicon","weight":0.181,"sharePct":67.57,"shareLowPct":27.028,"shareHighPct":100,"stale":false,"offBasis":false,"level":3.16666666667,"years":1.1224620483119663,"disagreement":false},{"id":"host-cpus","weight":0.039,"sharePct":39.88,"shareLowPct":29.91,"shareHighPct":54.84,"stale":false,"offBasis":false,"level":2.5,"years":0.7071067811865475,"disagreement":false},{"id":"processor-ip","weight":0.002,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":65.2777777778,"risingEligible":true,"risingComponents":{"growth":62.5,"shareChange":100,"committed":33.3333333333},"mc":{"medianRank":1,"rankP10":1,"rankP90":1,"pTop8":1,"pSelected":1,"stability":"firm"},"baselineRevenueRank":1},{"entityId":"broadcom","name":"Broadcom","reachPct":5.09663,"importance":5.67799548147,"importanceRank":2,"functions":[{"id":"merchant-accelerators","weight":0.711,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"custom-accelerators","weight":0.067,"sharePct":32.44,"shareLowPct":31.14,"shareHighPct":33.74,"stale":false,"offBasis":false,"level":3.5,"years":1.414213562373095,"disagreement":false},{"id":"networking-silicon","weight":0.181,"sharePct":16.15,"shareLowPct":6.46,"shareHighPct":25.84,"stale":false,"offBasis":false,"level":2.83333333333,"years":0.890898718138281,"disagreement":false},{"id":"host-cpus","weight":0.039,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"processor-ip","weight":0.002,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":47.2222222222,"risingEligible":true,"risingComponents":{"growth":50,"shareChange":25,"committed":66.6666666667},"mc":{"medianRank":2,"rankP10":2,"rankP90":2,"pTop8":1,"pSelected":1,"stability":"firm"},"baselineRevenueRank":2},{"entityId":"huawei","name":"Huawei","reachPct":3.104371,"importance":3.28571236662,"importanceRank":3,"functions":[{"id":"merchant-accelerators","weight":0.711,"sharePct":3.983,"shareLowPct":1.5932,"shareHighPct":6.3728,"stale":false,"offBasis":false,"level":3.16666666667,"years":1.1224620483119663,"disagreement":false},{"id":"custom-accelerators","weight":0.067,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"networking-silicon","weight":0.181,"sharePct":0.646,"shareLowPct":0.2153,"shareHighPct":2.153,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"host-cpus","weight":0.039,"sharePct":3.988,"shareLowPct":1.994,"shareHighPct":6.98,"stale":false,"offBasis":false,"level":2,"years":0.5,"disagreement":false},{"id":"processor-ip","weight":0.002,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{},"mc":{"medianRank":3,"rankP10":3,"rankP90":4,"pTop8":1,"pSelected":1,"stability":"firm"},"baselineRevenueRank":null},{"entityId":"amd","name":"AMD","reachPct":3.532375,"importance":2.12684527128,"importanceRank":4,"functions":[{"id":"merchant-accelerators","weight":0.711,"sharePct":3.699,"shareLowPct":1.4796,"shareHighPct":5.9184,"stale":false,"offBasis":false,"level":2.33333333333,"years":0.6299605249459811,"disagreement":false},{"id":"custom-accelerators","weight":0.067,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"networking-silicon","weight":0.181,"sharePct":0.646,"shareLowPct":0.2153,"shareHighPct":1.292,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"host-cpus","weight":0.039,"sharePct":20.14,"shareLowPct":8.056,"shareHighPct":32.224,"stale":false,"offBasis":false,"level":2.16666666667,"years":0.5612310241559831,"disagreement":false},{"id":"processor-ip","weight":0.002,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":12.5,"risingEligible":true,"risingComponents":{"growth":12.5,"shareChange":12.5},"mc":{"medianRank":4,"rankP10":3,"rankP90":5,"pTop8":1,"pSelected":1,"stability":"firm"},"baselineRevenueRank":4},{"entityId":"alphabet","name":"Alphabet","reachPct":1.5799506,"importance":1.72249972885,"importanceRank":5,"functions":[{"id":"merchant-accelerators","weight":0.711,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"custom-accelerators","weight":0.067,"sharePct":22.71,"shareLowPct":19.46,"shareHighPct":27.58,"stale":false,"offBasis":false,"level":3.16666666667,"years":1.1224620483119663,"disagreement":false},{"id":"networking-silicon","weight":0.181,"sharePct":0.1077,"shareLowPct":0,"shareHighPct":0.4307,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"host-cpus","weight":0.039,"sharePct":0.9971,"shareLowPct":0.1994,"shareHighPct":2.991,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"processor-ip","weight":0.002,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{},"mc":{"medianRank":5,"rankP10":4,"rankP90":5,"pTop8":1,"pSelected":1,"stability":"firm"},"baselineRevenueRank":null},{"entityId":"amazon","name":"Amazon","reachPct":1.2377727,"importance":1.02236913296,"importanceRank":6,"functions":[{"id":"merchant-accelerators","weight":0.711,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"custom-accelerators","weight":0.067,"sharePct":16.15,"shareLowPct":6.46,"shareHighPct":25.84,"stale":false,"offBasis":false,"level":2.83333333333,"years":0.890898718138281,"disagreement":false},{"id":"networking-silicon","weight":0.181,"sharePct":0.4307,"shareLowPct":0.1077,"shareHighPct":1.077,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"host-cpus","weight":0.039,"sharePct":1.994,"shareLowPct":0.4986,"shareHighPct":4.986,"stale":false,"offBasis":false,"level":2,"years":0.5,"disagreement":false},{"id":"processor-ip","weight":0.002,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{"committed":100},"mc":{"medianRank":6,"rankP10":6,"rankP90":8,"pTop8":0.971,"pSelected":0.971,"stability":"firm"},"baselineRevenueRank":null},{"entityId":"marvell","name":"Marvell","reachPct":1.430091,"importance":0.802611436566,"importanceRank":7,"functions":[{"id":"merchant-accelerators","weight":0.711,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"custom-accelerators","weight":0.067,"sharePct":3.893,"shareLowPct":1.5572,"shareHighPct":6.2288,"stale":false,"offBasis":false,"level":2.16666666667,"years":0.5612310241559831,"disagreement":false},{"id":"networking-silicon","weight":0.181,"sharePct":6.46,"shareLowPct":2.584,"shareHighPct":10.336,"stale":false,"offBasis":false,"level":2.16666666667,"years":0.5612310241559831,"disagreement":false},{"id":"host-cpus","weight":0.039,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"processor-ip","weight":0.002,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":31.25,"risingEligible":true,"risingComponents":{"growth":25,"shareChange":37.5},"mc":{"medianRank":7,"rankP10":6,"rankP90":9,"pTop8":0.8499,"pSelected":0.8499,"stability":"likely"},"baselineRevenueRank":5},{"entityId":"intel","name":"Intel","reachPct":1.3711672,"importance":0.705800790137,"importanceRank":8,"functions":[{"id":"merchant-accelerators","weight":0.711,"sharePct":0.1707,"shareLowPct":0.0569,"shareHighPct":0.2845,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerators","weight":0.067,"sharePct":0.9473,"shareLowPct":0.37892,"shareHighPct":1.51568,"stale":false,"offBasis":false,"level":2.33333333333,"years":0.6299605249459811,"disagreement":false},{"id":"networking-silicon","weight":0.181,"sharePct":0.5384,"shareLowPct":0.2153,"shareHighPct":1.077,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"host-cpus","weight":0.039,"sharePct":27.92,"shareLowPct":11.168,"shareHighPct":44.672,"stale":false,"offBasis":false,"level":2.16666666667,"years":0.5612310241559831,"disagreement":false},{"id":"processor-ip","weight":0.002,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":0,"risingEligible":true,"risingComponents":{"growth":0,"shareChange":0},"mc":{"medianRank":8,"rankP10":6,"rankP90":9,"pTop8":0.7882,"pSelected":0.7882,"stability":"likely"},"baselineRevenueRank":3},{"entityId":"arm","name":"Arm (SoftBank group: Arm, Ampere, Graphcore)","reachPct":0.205288195,"importance":0.468564423532,"importanceRank":9,"functions":[{"id":"merchant-accelerators","weight":0.711,"sharePct":0.002845,"shareLowPct":0,"shareHighPct":0.01138,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerators","weight":0.067,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"networking-silicon","weight":0.181,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"host-cpus","weight":0.039,"sharePct":0.4986,"shareLowPct":0.09971,"shareHighPct":1.994,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"processor-ip","weight":0.002,"sharePct":91.91,"shareLowPct":36.764,"shareHighPct":100,"stale":false,"offBasis":false,"level":4.33333333333,"years":2.519842099783924,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":10,"rankP10":8,"rankP90":11,"pTop8":0.2112,"pSelected":0.2112,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"alibaba","name":"Alibaba","reachPct":0.49378587,"importance":0.4218309675,"importanceRank":10,"functions":[{"id":"merchant-accelerators","weight":0.711,"sharePct":0.08877,"shareLowPct":0.02731,"shareHighPct":0.2845,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerators","weight":0.067,"sharePct":5.938,"shareLowPct":3.114,"shareHighPct":9.759,"stale":false,"offBasis":false,"level":3,"years":1,"disagreement":false},{"id":"networking-silicon","weight":0.181,"sharePct":0.0646,"shareLowPct":0,"shareHighPct":0.2153,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"host-cpus","weight":0.039,"sharePct":0.4986,"shareLowPct":0.09971,"shareHighPct":1.994,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"processor-ip","weight":0.002,"sharePct":0.8432,"shareLowPct":0,"shareHighPct":3.373,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true}],"rising":null,"risingEligible":true,"risingComponents":{},"mc":{"medianRank":9,"rankP10":8,"rankP90":11,"pTop8":0.1671,"pSelected":0.1671,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"credo","name":"Credo","reachPct":0.467704,"importance":0.330716669988,"importanceRank":11,"functions":[{"id":"merchant-accelerators","weight":0.711,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"custom-accelerators","weight":0.067,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"networking-silicon","weight":0.181,"sharePct":2.584,"shareLowPct":1.0336,"shareHighPct":4.1344,"stale":false,"offBasis":false,"level":2.5,"years":0.7071067811865475,"disagreement":false},{"id":"host-cpus","weight":0.039,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"processor-ip","weight":0.002,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":81.25,"risingEligible":true,"risingComponents":{"growth":87.5,"shareChange":75},"mc":{"medianRank":11,"rankP10":10,"rankP90":12,"pTop8":0.0126,"pSelected":1,"stability":"firm"},"baselineRevenueRank":7},{"entityId":"meta","name":"Meta","reachPct":0.3377146,"importance":0.208304140264,"importanceRank":12,"functions":[{"id":"merchant-accelerators","weight":0.711,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"custom-accelerators","weight":0.067,"sharePct":4.866,"shareLowPct":3.244,"shareHighPct":6.488,"stale":false,"offBasis":false,"level":2.33333333333,"years":0.6299605249459811,"disagreement":false},{"id":"networking-silicon","weight":0.181,"sharePct":0.0646,"shareLowPct":0,"shareHighPct":0.2153,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"host-cpus","weight":0.039,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"processor-ip","weight":0.002,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{},"mc":{"medianRank":12,"rankP10":11,"rankP90":13,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"bytedance","name":"ByteDance","reachPct":0.217348,"importance":0.172509221922,"importanceRank":13,"functions":[{"id":"merchant-accelerators","weight":0.711,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"custom-accelerators","weight":0.067,"sharePct":3.244,"shareLowPct":1.2976,"shareHighPct":5.1904,"stale":false,"offBasis":false,"level":2.66666666667,"years":0.7937005259859335,"disagreement":false},{"id":"networking-silicon","weight":0.181,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"host-cpus","weight":0.039,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"processor-ip","weight":0.002,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":13,"rankP10":12,"rankP90":16,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"astera-labs","name":"Astera Labs","reachPct":0.311863,"importance":0.1559315,"importanceRank":14,"functions":[{"id":"merchant-accelerators","weight":0.711,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"custom-accelerators","weight":0.067,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"networking-silicon","weight":0.181,"sharePct":1.723,"shareLowPct":0.6892,"shareHighPct":2.7568,"stale":false,"offBasis":false,"level":2,"years":0.5,"disagreement":false},{"id":"host-cpus","weight":0.039,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"processor-ip","weight":0.002,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":68.75,"risingEligible":true,"risingComponents":{"growth":75,"shareChange":62.5},"mc":{"medianRank":14,"rankP10":13,"rankP90":17,"pTop8":0,"pSelected":0.0126,"stability":"borderline"},"baselineRevenueRank":10},{"entityId":"hygon","name":"Hygon","reachPct":0.3111675,"importance":0.105013875,"importanceRank":15,"functions":[{"id":"merchant-accelerators","weight":0.711,"sharePct":0.2845,"shareLowPct":0.1707,"shareHighPct":0.4552,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerators","weight":0.067,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"networking-silicon","weight":0.181,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"host-cpus","weight":0.039,"sharePct":2.792,"shareLowPct":1.496,"shareHighPct":4.487,"stale":false,"offBasis":false,"level":2,"years":0.5,"disagreement":false},{"id":"processor-ip","weight":0.002,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":29.1666666667,"risingEligible":true,"risingComponents":{"growth":37.5,"shareChange":50,"committed":0},"mc":{"medianRank":15,"rankP10":14,"rankP90":17,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":6},{"entityId":"baidu","name":"Baidu","reachPct":0.18870853,"importance":0.0960860363773,"importanceRank":16,"functions":[{"id":"merchant-accelerators","weight":0.711,"sharePct":0.1526,"shareLowPct":0.1387,"shareHighPct":0.1941,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerators","weight":0.067,"sharePct":1.139,"shareLowPct":0.7589,"shareHighPct":1.265,"stale":false,"offBasis":false,"level":2.83333333333,"years":0.890898718138281,"disagreement":false},{"id":"networking-silicon","weight":0.181,"sharePct":0.02153,"shareLowPct":0,"shareHighPct":0.08614,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"host-cpus","weight":0.039,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"processor-ip","weight":0.002,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":17,"rankP10":15,"rankP90":18,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":12},{"entityId":"cambricon","name":"Cambricon","reachPct":0.3647082,"importance":0.09117705,"importanceRank":17,"functions":[{"id":"merchant-accelerators","weight":0.711,"sharePct":0.5128,"shareLowPct":0.48716,"shareHighPct":0.53844,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerators","weight":0.067,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"networking-silicon","weight":0.181,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"host-cpus","weight":0.039,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"processor-ip","weight":0.002,"sharePct":0.0537,"shareLowPct":0.051015,"shareHighPct":0.056385,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true}],"rising":93.75,"risingEligible":true,"risingComponents":{"growth":100,"shareChange":87.5},"mc":{"medianRank":16,"rankP10":14,"rankP90":18,"pTop8":0,"pSelected":1,"stability":"firm"},"baselineRevenueRank":9},{"entityId":"cisco","name":"Cisco","reachPct":0.226069,"importance":0.0897155421046,"importanceRank":18,"functions":[{"id":"merchant-accelerators","weight":0.711,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"custom-accelerators","weight":0.067,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"networking-silicon","weight":0.181,"sharePct":1.249,"shareLowPct":0.5017,"shareHighPct":2.5066,"stale":false,"offBasis":false,"level":1.66666666667,"years":0.3968502629929668,"disagreement":false},{"id":"host-cpus","weight":0.039,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"processor-ip","weight":0.002,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":18,"rankP10":14,"rankP90":19,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"alchip","name":"Alchip Technologies","reachPct":0.113029,"importance":0.0712038081741,"importanceRank":19,"functions":[{"id":"merchant-accelerators","weight":0.711,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"custom-accelerators","weight":0.067,"sharePct":1.687,"shareLowPct":0.6748,"shareHighPct":2.6992,"stale":false,"offBasis":false,"level":2.33333333333,"years":0.6299605249459811,"disagreement":false},{"id":"networking-silicon","weight":0.181,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"host-cpus","weight":0.039,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"processor-ip","weight":0.002,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":19,"rankP10":17,"rankP90":20,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":11},{"entityId":"cerebras","name":"Cerebras","reachPct":0.145044,"importance":0.036261,"importanceRank":20,"functions":[{"id":"merchant-accelerators","weight":0.711,"sharePct":0.204,"shareLowPct":0.0816,"shareHighPct":0.3264,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerators","weight":0.067,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"networking-silicon","weight":0.181,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"host-cpus","weight":0.039,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"processor-ip","weight":0.002,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":20,"rankP10":19,"rankP90":22,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":13},{"entityId":"guc","name":"GUC","reachPct":0.0407226,"importance":0.0256536304732,"importanceRank":21,"functions":[{"id":"merchant-accelerators","weight":0.711,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"custom-accelerators","weight":0.067,"sharePct":0.6078,"shareLowPct":0.24312,"shareHighPct":0.97248,"stale":false,"offBasis":false,"level":2.33333333333,"years":0.6299605249459811,"disagreement":false},{"id":"networking-silicon","weight":0.181,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"host-cpus","weight":0.039,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"processor-ip","weight":0.002,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":23,"rankP10":21,"rankP90":23,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":8},{"entityId":"groq","name":"Groq","reachPct":0.1011753,"importance":0.025293825,"importanceRank":22,"functions":[{"id":"merchant-accelerators","weight":0.711,"sharePct":0.1423,"shareLowPct":0,"shareHighPct":0.297275,"stale":false,"offBasis":true,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerators","weight":0.067,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"networking-silicon","weight":0.181,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"host-cpus","weight":0.039,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"processor-ip","weight":0.002,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":22,"rankP10":20,"rankP90":23,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"microsoft","name":"Microsoft","reachPct":0.0911053,"importance":0.022776325,"importanceRank":23,"functions":[{"id":"merchant-accelerators","weight":0.711,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"custom-accelerators","weight":0.067,"sharePct":0.7786,"shareLowPct":0.2595,"shareHighPct":1.557,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"networking-silicon","weight":0.181,"sharePct":0.1077,"shareLowPct":0.02153,"shareHighPct":0.323,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"host-cpus","weight":0.039,"sharePct":0.4986,"shareLowPct":0.09971,"shareHighPct":1.994,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"processor-ip","weight":0.002,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":21,"rankP10":20,"rankP90":22,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"tencent","name":"Tencent","reachPct":0.02789567,"importance":0.0069739175,"importanceRank":24,"functions":[{"id":"merchant-accelerators","weight":0.711,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"custom-accelerators","weight":0.067,"sharePct":0.3,"shareLowPct":0.15,"shareHighPct":0.7501,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"networking-silicon","weight":0.181,"sharePct":0.04307,"shareLowPct":0,"shareHighPct":0.1292,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"host-cpus","weight":0.039,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"processor-ip","weight":0.002,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":24,"rankP10":23,"rankP90":24,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null}],"selection":{"importanceIds":["nvidia","broadcom","huawei","amd","alphabet","amazon","marvell","intel"],"risingIds":["cambricon","credo"],"filledFromImportance":[]},"risingPools":{"growth":9,"shareChange":9,"committed":4},"weightSensitivity":[{"functionId":"merchant-accelerators","end":"low","entered":[],"left":[]},{"functionId":"merchant-accelerators","end":"high","entered":[],"left":[]},{"functionId":"custom-accelerators","end":"low","entered":[],"left":[]},{"functionId":"custom-accelerators","end":"high","entered":[],"left":[]},{"functionId":"networking-silicon","end":"low","entered":[],"left":[]},{"functionId":"networking-silicon","end":"high","entered":[],"left":[]},{"functionId":"host-cpus","end":"low","entered":[],"left":[]},{"functionId":"host-cpus","end":"high","entered":[],"left":[]},{"functionId":"processor-ip","end":"low","entered":[],"left":[]},{"functionId":"processor-ip","end":"high","entered":["arm"],"left":["intel"]}],"validation":[{"caseId":"chips-chip-design-nvidia-accelerator-allocation-2023","entityId":"nvidia","rank":1,"pass":true}],"baseline":{"spearman":0.846153846154,"baselineTop10NotSelected":["hygon","guc","astera-labs"],"selectedNotInBaselineTop10":["huawei","alphabet","amazon"]},"raterAgreement":{"within":0.851851851852,"weightedKappa":0.817813765182,"disagreements":[]}},"inputs":{"nvidia":{"schema":"5lc-index-inputs/1","key":"chips-chip-design","entityId":"nvidia","name":"NVIDIA","kind":"public","commercialActivityVerified":true,"shares":{"merchant-accelerators":{"low":89.26,"central":90.11,"high":90.68,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2026 (to 25 January 2026)","periodEnd":"2026-01-25","source":{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","title":"NVIDIA Corporation Form 10-K for the fiscal year ended January 25, 2026","publisher":"NVIDIA / U.S. SEC","publishedDate":"2026-02-25","accessedDate":"2026-10-22","tier":"T5","locator":"Notes, revenue by specialized markets: Data Center Compute $162,361M"},"note":"T5 derivation (full inputs). NVIDIA Data Center Compute $162.361B in fiscal 2026 (to 25 January 2026) less an estimated $4.0B (range $3.0–5.5B) of Grace processors in GB200/GB300, which the card counts as host processors. 2025 value about $158.4B. Function total $175.74B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $168.7–183.3B (central $175.8B)). Share = 158.4 / 175.74 = 90.11%. Value range $157-159B gives the share range 89.26-90.68%. Itemisation and 2024 prior in a local evidence capture."},"custom-accelerators":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","title":"NVIDIA Corporation Form 10-K for the fiscal year ended January 25, 2026","publisher":"NVIDIA / U.S. SEC","publishedDate":"2026-02-25","accessedDate":"2026-10-22","tier":"T1","locator":"Item 1 Business and revenue by specialized markets: NVIDIA sells its own GPUs, CPUs, DPUs and networking platforms; no custom accelerator design service for other companies' chips is reported in fiscal 2026"}},"networking-silicon":{"low":67.57,"central":67.57,"high":67.57,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2026 (to 25 January 2026)","periodEnd":"2026-01-25","source":{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","title":"NVIDIA Corporation Form 10-K for the fiscal year ended January 25, 2026","publisher":"NVIDIA / U.S. SEC","publishedDate":"2026-02-25","accessedDate":"2026-10-22","tier":"T5","locator":"Notes, revenue by specialized markets: Data Center Networking $31,376M"},"note":"T5 derivation (full inputs). NVIDIA Data Center Networking $31.376B in fiscal 2026 (switches, adapters, NVLink and cables, built on its own chips; card counts it at system level). 2025 value about $31.38B. Function total $46.44B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $35.1–47.1B (central $44.8B)). Share = 31.38 / 46.44 = 67.57%. Point value: the engine applies the T5 width. Itemisation and 2024 prior in a local evidence capture."},"host-cpus":{"low":29.91,"central":39.88,"high":54.84,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2026 (to 25 January 2026)","periodEnd":"2026-01-25","source":{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","title":"NVIDIA Corporation Form 10-K for the fiscal year ended January 25, 2026","publisher":"NVIDIA / U.S. SEC","publishedDate":"2026-02-25","accessedDate":"2026-10-22","tier":"T5","locator":"Item 1 Business: Grace CPU in GB200/GB300 systems"},"note":"T5 derivation (full inputs). Grace CPUs inside GB200 and GB300 modules, estimated at $4.0B (range $3.0–5.5B; card T5 estimate, consistent with Dell'Oro's September 2025 report that Arm CPUs reached a quarter of server-processor spending on GB200/GB300 and cloud chips). NVIDIA does not report Grace revenue. 2025 value about $4B. Function total $10.03B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $7.0–12.9B (central $9.6B)). Share = 4 / 10.03 = 39.88%. Value range $3-5.5B gives the share range 29.91-54.84%. Itemisation and 2024 prior in a local evidence capture."},"processor-ip":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","title":"NVIDIA Corporation Form 10-K for the fiscal year ended January 25, 2026","publisher":"NVIDIA / U.S. SEC","publishedDate":"2026-02-25","accessedDate":"2026-10-22","tier":"T1","locator":"Item 1 Business: NVIDIA sells chips, systems and software; it reports no processor-core licensing revenue"}}},"priorShares":{"merchant-accelerators":{"low":90.2,"central":90.65,"high":90.91,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2025 (to 26 January 2025)","periodEnd":"2025-01-26","source":{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","title":"NVIDIA Corporation Form 10-K for the fiscal year ended January 25, 2026","publisher":"NVIDIA / U.S. SEC","publishedDate":"2026-02-25","accessedDate":"2026-10-22","tier":"T5","locator":"Notes, revenue by specialized markets, fiscal 2025 column: Data Center Compute $102,196M"},"note":"T5 derivation of the prior-year share. NVIDIA Data Center Compute $102.196B in fiscal 2025 (to 26 January 2025) less an estimated $0.7B (range $0.4-1.2B) of Grace processors (GH200 and first GB200 shipments). 2024 value about $101.5B. Function total $111.97B = the itemised 2024 values of every candidate plus residual others, built like the 2025 total; candidates without a reported 2024 value are set at half their 2025 value (an assumption: most grew at least 100%). Share = 101.5 / 111.97 = 90.65%. Value range $101-102B gives the share range 90.2-90.91%. Itemisation in a local evidence capture."},"networking-silicon":{"low":60.92,"central":60.92,"high":60.92,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2025 (to 26 January 2025)","periodEnd":"2025-01-26","source":{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","title":"NVIDIA Corporation Form 10-K for the fiscal year ended January 25, 2026","publisher":"NVIDIA / U.S. SEC","publishedDate":"2026-02-25","accessedDate":"2026-10-22","tier":"T5","locator":"Notes, revenue by specialized markets, fiscal 2025 column: Networking $12,990M"},"note":"T5 derivation of the prior-year share. NVIDIA Data Center Networking $12.990B in fiscal 2025. 2024 value about $12.99B. Function total $21.32B = the itemised 2024 values of every candidate plus residual others, built like the 2025 total; candidates without a reported 2024 value are set at half their 2025 value (an assumption: most grew at least 100%). Share = 12.99 / 21.32 = 60.92%. Point value: the engine applies the T5 width. Itemisation in a local evidence capture."},"host-cpus":{"low":7.262,"central":12.71,"high":21.79,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2025 (to 26 January 2025)","periodEnd":"2025-01-26","source":{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","title":"NVIDIA Corporation Form 10-K for the fiscal year ended January 25, 2026","publisher":"NVIDIA / U.S. SEC","publishedDate":"2026-02-25","accessedDate":"2026-10-22","tier":"T5","locator":"Item 1 Business and fiscal 2025 compute revenue: Grace in GH200 and early GB200"},"note":"T5 derivation of the prior-year share. Grace processors in GH200 and the first GB200 shipments, estimated at $0.7B (range $0.4-1.2B); NVIDIA does not report Grace revenue. 2024 value about $0.7B. Function total $5.51B = the itemised 2024 values of every candidate plus residual others, built like the 2025 total; candidates without a reported 2024 value are set at half their 2025 value (an assumption: most grew at least 100%). Share = 0.7 / 5.51 = 12.71%. Value range $0.4-1.2B gives the share range 7.262-21.79%. Itemisation in a local evidence capture."}},"activityGrowth":{"current":{"low":193737000000,"central":193737000000,"high":193737000000,"unit":"USD","basis":"Data Center revenue (compute plus networking), fiscal year, US dollars","period":"FY2026 (to 25 January 2026)","periodEnd":"2026-01-25","source":{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","title":"NVIDIA Corporation Form 10-K for the fiscal year ended January 25, 2026","publisher":"NVIDIA / U.S. SEC","publishedDate":"2026-02-25","accessedDate":"2026-10-22","tier":"T1","locator":"Notes, revenue by specialized markets: Data Center $193,737M"},"note":"Data Center revenue is NVIDIA's AI data-centre chip and system activity: it covers all three parts NVIDIA participates in (accelerators, networking, Grace host processors)."},"prior":{"low":115186000000,"central":115186000000,"high":115186000000,"unit":"USD","basis":"Data Center revenue (compute plus networking), fiscal year, US dollars","period":"FY2025 (to 26 January 2025)","periodEnd":"2025-01-26","source":{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","title":"NVIDIA Corporation Form 10-K for the fiscal year ended January 25, 2026","publisher":"NVIDIA / U.S. SEC","publishedDate":"2026-02-25","accessedDate":"2026-10-22","tier":"T1","locator":"Notes, revenue by specialized markets, fiscal 2025 column: Data Center $115,186M"}}},"committedExpansion":{"low":0.4914,"central":0.4914,"high":0.4914,"unit":"ratio","basis":"Inventory purchase and long-term supply and capacity obligations ÷ fiscal 2026 Data Center revenue","period":"FY2026 (to 25 January 2026)","periodEnd":"2026-01-25","source":{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","title":"NVIDIA Corporation Form 10-K for the fiscal year ended January 25, 2026","publisher":"NVIDIA / U.S. SEC","publishedDate":"2026-02-25","accessedDate":"2026-10-22","tier":"T1","locator":"Auditor's critical audit matter: outstanding inventory purchase and long-term supply and capacity obligations balance of $95.2 billion at 25 January 2026; Note 9 RPO $2.3 billion"},"note":"Note on inventory and supply commitments / MD&A: outstanding inventory purchase and long-term supply and capacity obligations of $95.2B at 25 January 2026 (critical audit matter); Data Center revenue $193.737B; 95.2 / 193.737 = 0.4914. These are NVIDIA's own commitments to suppliers for future production, the closest contracted measure NVIDIA reports: its remaining performance obligations are only $2.3B (contracts over a year), and management's $500B Blackwell and Rubin 'visibility' through 2026 is not a contract, so it is not used."},"baselineRevenueUsd":{"low":193737000000,"central":193737000000,"high":193737000000,"unit":"USD","basis":"Data Center revenue (compute plus networking), fiscal year, US dollars","period":"FY2026 (to 25 January 2026)","periodEnd":"2026-01-25","source":{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","title":"NVIDIA Corporation Form 10-K for the fiscal year ended January 25, 2026","publisher":"NVIDIA / U.S. SEC","publishedDate":"2026-02-25","accessedDate":"2026-10-22","tier":"T1","locator":"Notes, revenue by specialized markets: Data Center $193,737M"},"note":"Data Center revenue only, never NVIDIA's total ($215.9B)."},"screening":{"merchant-accelerators":{"levelLow":3,"levelHigh":5},"networking-silicon":{"levelLow":2,"levelHigh":4},"host-cpus":{"levelLow":2,"levelHigh":4}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"d5048cce5bf090d988ab78c5be3b9186ae86d70b1779322cc63136e602d506ec"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"d5048cce5bf090d988ab78c5be3b9186ae86d70b1779322cc63136e602d506ec"}}},"broadcom":{"schema":"5lc-index-inputs/1","key":"chips-chip-design","entityId":"broadcom","name":"Broadcom","kind":"public","commercialActivityVerified":true,"shares":{"merchant-accelerators":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/1730168/000173016825000121/avgo-20251102.htm","title":"Broadcom Inc. Form 10-K for the fiscal year ended November 2, 2025","publisher":"Broadcom / U.S. SEC","publishedDate":"2025-12-18","accessedDate":"2026-10-22","tier":"T1","locator":"Item 1 Business: Broadcom sells custom accelerators (XPUs) to their owners and networking products; it sells no accelerators of its own design on the open market"}},"custom-accelerators":{"low":31.14,"central":32.44,"high":33.74,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2025 (to 2 November 2025)","periodEnd":"2025-11-02","source":{"url":"https://www.fool.com/earnings/call-transcripts/2025/12/12/broadcom-avgo-q4-2025-earnings-call-transcript/","title":"Broadcom (AVGO) Q4 2025 earnings call transcript","publisher":"The Motley Fool (transcript of Broadcom's earnings call)","publishedDate":"2025-12-12","accessedDate":"2026-10-22","tier":"T5","locator":"Hock Tan: AI revenue grew 65% year over year to $20 billion"},"note":"T5 derivation (full inputs). Broadcom fiscal 2025 AI revenue $20B (Q4 call); XPUs were 65% of AI revenue on the Q3 FY2025 call, so XPUs are about $12.5B (60–65%). Its programmes (Google, Meta, ByteDance and others) total $12.5B; Broadcom's design-partner half is $6.25B. 2025 value about $6.25B. Function total $19.27B (was $17.03B before Amazon's programme was re-derived from chip volume, a recorded method decision; see fixes/a recorded method decision/itemisation-changes-a recorded method decision.md) = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $14.7–20.5B (central $16.5B)). Share = 6.25 / 19.27 = 32.44%. Value range $6-6.5B gives the share range 31.14-33.74%. Itemisation and 2024 prior in a local evidence capture."},"networking-silicon":{"low":16.15,"central":16.15,"high":16.15,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2025 (to 2 November 2025)","periodEnd":"2025-11-02","source":{"url":"https://www.fool.com/earnings/call-transcripts/2025/12/12/broadcom-avgo-q4-2025-earnings-call-transcript/","title":"Broadcom (AVGO) Q4 2025 earnings call transcript","publisher":"The Motley Fool (transcript of Broadcom's earnings call)","publishedDate":"2025-12-12","accessedDate":"2026-10-22","tier":"T5","locator":"Hock Tan: AI switch backlog above $10 billion; AI revenue $20 billion"},"note":"T5 derivation (full inputs). AI networking is the remaining 35–40% of $20B: about $7.5B (Tomahawk and Jericho switch chips, NICs, PCIe switches). 2025 value about $7.5B. Function total $46.44B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $35.1–47.1B (central $44.8B)). Share = 7.5 / 46.44 = 16.15%. Point value: the engine applies the T5 width. Itemisation and 2024 prior in a local evidence capture."},"host-cpus":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/1730168/000173016825000121/avgo-20251102.htm","title":"Broadcom Inc. Form 10-K for the fiscal year ended November 2, 2025","publisher":"Broadcom / U.S. SEC","publishedDate":"2025-12-18","accessedDate":"2026-10-22","tier":"T1","locator":"Item 1 Business: no server processors"}},"processor-ip":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/1730168/000173016825000121/avgo-20251102.htm","title":"Broadcom Inc. Form 10-K for the fiscal year ended November 2, 2025","publisher":"Broadcom / U.S. SEC","publishedDate":"2025-12-18","accessedDate":"2026-10-22","tier":"T1","locator":"Item 1 Business: no processor-core licensing"}}},"priorShares":{"custom-accelerators":{"low":25.47,"central":30.1,"high":34.73,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2024 (to 3 November 2024)","periodEnd":"2024-11-03","source":{"url":"https://www.constellationr.com/insights/news/broadcom-continues-ride-ai-infrastructure-wave-strong-q4","title":"Broadcom continues to ride AI infrastructure wave with strong Q4","publisher":"Constellation Research (Larry Dignan)","publishedDate":"2024-12-12","accessedDate":"2026-10-22","tier":"T5","locator":"Hock Tan: AI revenue $12.2 billion in fiscal 2024, up 220%"},"note":"T5 derivation of the prior-year share. Broadcom fiscal 2024 AI revenue $12.2B (company figure as reported by Constellation Research); XPUs taken as 65% (range 55-75%; Broadcom gave 65% for Q3 FY2025 and no split for fiscal 2024), so about $7.9B. Broadcom's design-partner half of its programmes is about $3.97B. 2024 value about $3.965B. Function total $13.17B (was $11.21B before Amazon's 2024 programme was restated with the same in-house multiple, a recorded method decision) = the itemised 2024 values of every candidate plus residual others, built like the 2025 total; candidates without a reported 2024 value are set at half their 2025 value (an assumption: most grew at least 100%). Share = 3.965 / 13.174 = 30.1%. Value range $3.35-4.58B gives the share range 25.47-34.73%. Itemisation in a local evidence capture."},"networking-silicon":{"low":14.3,"central":20.03,"high":25.75,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2024 (to 3 November 2024)","periodEnd":"2024-11-03","source":{"url":"https://www.constellationr.com/insights/news/broadcom-continues-ride-ai-infrastructure-wave-strong-q4","title":"Broadcom continues to ride AI infrastructure wave with strong Q4","publisher":"Constellation Research (Larry Dignan)","publishedDate":"2024-12-12","accessedDate":"2026-10-22","tier":"T5","locator":"Hock Tan: AI revenue $12.2 billion in fiscal 2024, up 220%"},"note":"T5 derivation of the prior-year share. AI networking is the remaining 35% (range 25-45%) of $12.2B: about $4.27B (Tomahawk and Jericho switch chips, NICs, PCIe switches). 2024 value about $4.27B. Function total $21.32B = the itemised 2024 values of every candidate plus residual others, built like the 2025 total; candidates without a reported 2024 value are set at half their 2025 value (an assumption: most grew at least 100%). Share = 4.27 / 21.32 = 20.03%. Value range $3.05-5.49B gives the share range 14.3-25.75%. Itemisation in a local evidence capture."}},"activityGrowth":{"current":{"low":20000000000,"central":20000000000,"high":20000000000,"unit":"USD","basis":"AI semiconductor revenue (custom accelerators plus AI networking), fiscal year, US dollars","period":"FY2025 (to 2 November 2025)","periodEnd":"2025-11-02","source":{"url":"https://www.fool.com/earnings/call-transcripts/2025/12/12/broadcom-avgo-q4-2025-earnings-call-transcript/","title":"Broadcom (AVGO) Q4 2025 earnings call transcript","publisher":"The Motley Fool (transcript of Broadcom's earnings call)","publishedDate":"2025-12-12","accessedDate":"2026-10-22","tier":"T2","locator":"Hock Tan: AI revenue grew 65% to about $20 billion"},"note":"Both are Broadcom's own AI revenue figures (company says); the fiscal 2025 figure is rounded to $20B in the call."},"prior":{"low":12200000000,"central":12200000000,"high":12200000000,"unit":"USD","basis":"AI semiconductor revenue (custom accelerators plus AI networking), fiscal year, US dollars","period":"FY2024 (to 3 November 2024)","periodEnd":"2024-11-03","source":{"url":"https://www.constellationr.com/insights/news/broadcom-continues-ride-ai-infrastructure-wave-strong-q4","title":"Broadcom continues to ride AI infrastructure wave with strong Q4","publisher":"Constellation Research (Larry Dignan)","publishedDate":"2024-12-12","accessedDate":"2026-10-22","tier":"T2","locator":"AI revenue $12.2 billion, up 220%"}}},"committedExpansion":{"low":3.65,"central":3.65,"high":3.65,"unit":"ratio","basis":"AI order backlog ÷ fiscal 2025 AI semiconductor revenue","period":"Q4 FY2025 call (11 December 2025)","periodEnd":"2025-11-02","source":{"url":"https://www.fool.com/earnings/call-transcripts/2025/12/12/broadcom-avgo-q4-2025-earnings-call-transcript/","title":"Broadcom (AVGO) Q4 2025 earnings call transcript","publisher":"The Motley Fool (transcript of Broadcom's earnings call)","publishedDate":"2025-12-12","accessedDate":"2026-10-22","tier":"T2","locator":"Hock Tan: total AI order backlog in excess of $73 billion, deliverable over roughly the next 18 months (Q4 FY2025 call, 11 December 2025)"},"note":"Hock Tan said total AI orders on hand exceeded $73B, to be delivered over roughly the next 18 months (six quarters in the Q&A); $73B / $20.0B (fiscal 2025 AI revenue) = 3.65. The backlog includes XPUs and about $20B of networking and other products."},"baselineRevenueUsd":{"low":20000000000,"central":20000000000,"high":20000000000,"unit":"USD","basis":"AI semiconductor revenue, fiscal year, US dollars","period":"FY2025 (to 2 November 2025)","periodEnd":"2025-11-02","source":{"url":"https://www.fool.com/earnings/call-transcripts/2025/12/12/broadcom-avgo-q4-2025-earnings-call-transcript/","title":"Broadcom (AVGO) Q4 2025 earnings call transcript","publisher":"The Motley Fool (transcript of Broadcom's earnings call)","publishedDate":"2025-12-12","accessedDate":"2026-10-22","tier":"T2","locator":"Hock Tan: AI revenue about $20 billion"},"note":"AI semiconductor revenue only (company says), never Broadcom's total ($63.9B)."},"screening":{"custom-accelerators":{"levelLow":2,"levelHigh":4},"networking-silicon":{"levelLow":2,"levelHigh":4}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"da79300252e27513556439606f4a30ba3b2381abef054bb3a6a41b584fef52fb"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"da79300252e27513556439606f4a30ba3b2381abef054bb3a6a41b584fef52fb"}}},"huawei":{"schema":"5lc-index-inputs/1","key":"chips-chip-design","entityId":"huawei","name":"Huawei","kind":"private","commercialActivityVerified":true,"shares":{"merchant-accelerators":{"low":3.983,"central":3.983,"high":3.983,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://kfgo.com/?p=1285523","title":"Chinese chipmakers claim nearly half of local market as Nvidia's lead shrinks (IDC data)","publisher":"Reuters (Che Pan and Laurie Chen), via KFGO","publishedDate":"2026-04-01","accessedDate":"2026-10-22","tier":"T5","locator":"IDC 2025 China accelerator shipments: Huawei about 812,000"},"note":"T5 derivation (full inputs). Ascend accelerators: IDC counts about 812,000 Huawei cards shipped in China in 2025 (Reuters); at about $8,600 a card that is $7.0B (card range $4.0–10.5B; Mizuho had projected about 700,000 units). Huawei sells them mostly inside Atlas servers and SuperPoDs (more than 300 Atlas 900 A3 SuperPoDs deployed by September 2025, https://www.huawei.com/en/news/2025/9/hc-xu-keynote-speech). 2025 value about $7B. Function total $175.74B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $168.7–183.3B (central $175.8B)). Share = 7 / 175.74 = 3.983%. Point value: the engine applies the T5 width. Itemisation and 2024 prior in a local evidence capture."},"custom-accelerators":{"participates":false,"evidence":{"url":"https://www.huawei.com/en/news/2025/9/hc-xu-keynote-speech","title":"Huawei Connect 2025: Eric Xu keynote on SuperPoD interconnect and Ascend roadmap","publisher":"Huawei","publishedDate":"2025-09-18","accessedDate":"2026-10-22","tier":"T2","locator":"Keynote: Huawei designs and sells its own Ascend and Kunpeng chips and systems; no design work on other companies' accelerators is described"}},"networking-silicon":{"low":0.2153,"central":0.646,"high":2.153,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.huawei.com/en/news/2025/9/hc-xu-keynote-speech","title":"Huawei Connect 2025: Eric Xu keynote on SuperPoD interconnect and Ascend roadmap","publisher":"Huawei","publishedDate":"2025-09-18","accessedDate":"2026-10-22","tier":"T5","locator":"SuperPoD interconnect (keynote theme)"},"note":"T5 derivation (full inputs). Huawei's own interconnect and NIC chips inside Ascend clusters and its switches; chip-level AI value estimated at $0.1–1.0B (central $0.3B). 2025 value about $0.3B. Function total $46.44B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $35.1–47.1B (central $44.8B)). Share = 0.3 / 46.44 = 0.646%. Value range $0.1-1B gives the share range 0.2153-2.153%. Itemisation and 2024 prior in a local evidence capture."},"host-cpus":{"low":1.994,"central":3.988,"high":6.98,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.huawei.com/en/news/2025/9/hc-xu-keynote-speech","title":"Huawei Connect 2025: Eric Xu keynote on SuperPoD interconnect and Ascend roadmap","publisher":"Huawei","publishedDate":"2025-09-18","accessedDate":"2026-10-22","tier":"T5","locator":"Atlas 900 A3 SuperPoD and Kunpeng processors"},"note":"T5 derivation (full inputs). Kunpeng processors host Ascend servers (about one Kunpeng per two Ascend NPUs in SuperPoDs): about 406,000 processors at $500–1,700 each gives $0.2–0.7B (central $0.4B). Huawei reports no Kunpeng revenue. 2025 value about $0.4B. Function total $10.03B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $7.0–12.9B (central $9.6B)). Share = 0.4 / 10.03 = 3.988%. Value range $0.2-0.7B gives the share range 1.994-6.98%. Itemisation and 2024 prior in a local evidence capture."},"processor-ip":{"participates":false,"evidence":{"url":"https://www.huawei.com/en/news/2025/9/hc-xu-keynote-speech","title":"Huawei Connect 2025: Eric Xu keynote on SuperPoD interconnect and Ascend roadmap","publisher":"Huawei","publishedDate":"2025-09-18","accessedDate":"2026-10-22","tier":"T2","locator":"Keynote: Huawei's monetisation of AI is hardware-focused; it sells chips and systems and describes no processor-core licensing"}}},"screening":{"merchant-accelerators":{"levelLow":2,"levelHigh":4},"networking-silicon":{"levelLow":1,"levelHigh":3},"host-cpus":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"06ed4f37f179c8301847f8724fe106f25d98ec714684eef530018847e753860f"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"06ed4f37f179c8301847f8724fe106f25d98ec714684eef530018847e753860f"}}},"amd":{"schema":"5lc-index-inputs/1","key":"chips-chip-design","entityId":"amd","name":"AMD","kind":"public","commercialActivityVerified":true,"shares":{"merchant-accelerators":{"low":3.699,"central":3.699,"high":3.699,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2025 (to 27 December 2025)","periodEnd":"2025-12-27","source":{"url":"https://www.sec.gov/Archives/edgar/data/2488/000000248826000018/amd-20251227.htm","title":"Advanced Micro Devices Form 10-K for the fiscal year ended December 27, 2025","publisher":"AMD / U.S. SEC","publishedDate":"2026-02-04","accessedDate":"2026-10-22","tier":"T5","locator":"MD&A: Data Center net revenue of $16.6 billion"},"note":"T5 derivation (full inputs). AMD Instinct GPUs estimated at $6.5B: Data Center segment $16.6B in 2025 less an estimated $10.1B of EPYC server processors; AMD does not report Instinct revenue (The Next Platform estimated Q4 2025 Instinct at about $2.65B, https://www.nextplatform.com/2026/02/04/amd-finally-makes-more-money-on-gpus-than-cpus-in-a-quarter). 2025 value about $6.5B. Function total $175.74B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $168.7–183.3B (central $175.8B)). Share = 6.5 / 175.74 = 3.699%. Point value: the engine applies the T5 width. Itemisation and 2024 prior in a local evidence capture."},"custom-accelerators":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/2488/000000248826000018/amd-20251227.htm","title":"Advanced Micro Devices Form 10-K for the fiscal year ended December 27, 2025","publisher":"AMD / U.S. SEC","publishedDate":"2026-02-04","accessedDate":"2026-10-22","tier":"T1","locator":"Item 1 Business: Data Center products are EPYC CPUs, Instinct GPUs, Pensando DPUs/NICs and adaptive SoCs; semi-custom work is for game consoles, not data-centre AI accelerators"}},"networking-silicon":{"low":0.2153,"central":0.646,"high":1.292,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2025 (to 27 December 2025)","periodEnd":"2025-12-27","source":{"url":"https://www.sec.gov/Archives/edgar/data/2488/000000248826000018/amd-20251227.htm","title":"Advanced Micro Devices Form 10-K for the fiscal year ended December 27, 2025","publisher":"AMD / U.S. SEC","publishedDate":"2026-02-04","accessedDate":"2026-10-22","tier":"T5","locator":"Item 1: AMD Pensando DPUs, Pollara 400 and Vulcano AI NICs"},"note":"T5 derivation (full inputs). Pensando DPUs and Pollara/Vulcano AI NICs; AMD reports no networking revenue, so the AI part is estimated at $0.1–0.6B. 2025 value about $0.3B. Function total $46.44B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $35.1–47.1B (central $44.8B)). Share = 0.3 / 46.44 = 0.646%. Value range $0.1-0.6B gives the share range 0.2153-1.292%. Itemisation and 2024 prior in a local evidence capture."},"host-cpus":{"low":20.14,"central":20.14,"high":20.14,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2025 (to 27 December 2025)","periodEnd":"2025-12-27","source":{"url":"https://www.sec.gov/Archives/edgar/data/2488/000000248826000018/amd-20251227.htm","title":"Advanced Micro Devices Form 10-K for the fiscal year ended December 27, 2025","publisher":"AMD / U.S. SEC","publishedDate":"2026-02-04","accessedDate":"2026-10-22","tier":"T5","locator":"MD&A: Data Center net revenue of $16.6 billion"},"note":"T5 derivation (full inputs). EPYC server processors estimated at $10.1B; 20% of that ($2.02B) taken as AI-server hosts. 2025 value about $2.02B. Function total $10.03B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $7.0–12.9B (central $9.6B)). Share = 2.02 / 10.03 = 20.14%. Point value: the engine applies the T5 width. Itemisation and 2024 prior in a local evidence capture."},"processor-ip":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/2488/000000248826000018/amd-20251227.htm","title":"Advanced Micro Devices Form 10-K for the fiscal year ended December 27, 2025","publisher":"AMD / U.S. SEC","publishedDate":"2026-02-04","accessedDate":"2026-10-22","tier":"T1","locator":"Item 1 Business: AMD sells processors; it reports no processor-core licensing business"}}},"priorShares":{"merchant-accelerators":{"low":4.465,"central":4.644,"high":5.001,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2024 (to 28 December 2024)","periodEnd":"2024-12-28","source":{"url":"https://ir.amd.com/news-events/press-releases/detail/1236/amd-unveils-next-generation-ai-roadmap","title":"AMD Reports Fourth Quarter and Full Year 2024 Financial Results (the page URL slug reads amd-unveils-next-generation-ai-roadmap)","publisher":"AMD","publishedDate":"2025-02-04","accessedDate":"2026-10-22","tier":"T5","locator":"CEO: delivered more than $5 billion of AMD Instinct accelerator revenue in 2024"},"note":"T5 derivation of the prior-year share. AMD says it delivered more than $5B of Instinct accelerator revenue in 2024; taken as $5.2B (range $5.0-5.6B). 2024 value about $5.2B. Function total $111.97B = the itemised 2024 values of every candidate plus residual others, built like the 2025 total; candidates without a reported 2024 value are set at half their 2025 value (an assumption: most grew at least 100%). Share = 5.2 / 111.97 = 4.644%. Value range $5-5.6B gives the share range 4.465-5.001%. Itemisation in a local evidence capture."},"host-cpus":{"low":25.42,"central":26.87,"high":27.6,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2024 (to 28 December 2024)","periodEnd":"2024-12-28","source":{"url":"https://www.sec.gov/Archives/edgar/data/2488/000000248826000018/amd-20251227.htm","title":"Advanced Micro Devices Form 10-K for the fiscal year ended December 27, 2025","publisher":"AMD / U.S. SEC","publishedDate":"2026-02-04","accessedDate":"2026-10-22","tier":"T5","locator":"MD&A: Data Center net revenue of $12.6 billion in 2024"},"note":"T5 derivation of the prior-year share. EPYC server processors estimated at $7.4B (Data Center $12.6B less Instinct $5.2B); 20% ($1.48B, range $1.40-1.52B) taken as AI-server hosts, the same method as 2025. 2024 value about $1.48B. Function total $5.51B = the itemised 2024 values of every candidate plus residual others, built like the 2025 total; candidates without a reported 2024 value are set at half their 2025 value (an assumption: most grew at least 100%). Share = 1.48 / 5.51 = 26.87%. Value range $1.4-1.52B gives the share range 25.42-27.6%. Itemisation in a local evidence capture."},"networking-silicon":{"low":0.2345,"central":0.7035,"high":1.876,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2024 (to 28 December 2024)","periodEnd":"2024-12-28","source":{"url":"https://www.sec.gov/Archives/edgar/data/2488/000000248826000018/amd-20251227.htm","title":"Advanced Micro Devices Form 10-K for the fiscal year ended December 27, 2025","publisher":"AMD / U.S. SEC","publishedDate":"2026-02-04","accessedDate":"2026-10-22","tier":"T5","locator":"Item 1: Pensando DPUs and AI NICs"},"note":"T5 derivation of the prior-year share. Pensando DPUs and AI NICs in AI clusters; AMD reports no networking revenue, estimated at $0.05-0.4B. 2024 value about $0.15B. Function total $21.32B = the itemised 2024 values of every candidate plus residual others, built like the 2025 total; candidates without a reported 2024 value are set at half their 2025 value (an assumption: most grew at least 100%). Share = 0.15 / 21.32 = 0.7035%. Value range $0.05-0.4B gives the share range 0.2345-1.876%. Itemisation in a local evidence capture."}},"activityGrowth":{"current":{"low":16600000000,"central":16600000000,"high":16600000000,"unit":"USD","basis":"Data Center segment revenue, fiscal year, US dollars","period":"FY2025 (to 27 December 2025)","periodEnd":"2025-12-27","source":{"url":"https://www.sec.gov/Archives/edgar/data/2488/000000248826000018/amd-20251227.htm","title":"Advanced Micro Devices Form 10-K for the fiscal year ended December 27, 2025","publisher":"AMD / U.S. SEC","publishedDate":"2026-02-04","accessedDate":"2026-10-22","tier":"T1","locator":"MD&A: Data Center net revenue of $16.6 billion in 2025"},"note":"AMD reports no Instinct revenue, so the Data Center segment (EPYC and Instinct together) is the closest reported measure; it includes EPYC sold to servers that are not AI servers."},"prior":{"low":12600000000,"central":12600000000,"high":12600000000,"unit":"USD","basis":"Data Center segment revenue, fiscal year, US dollars","period":"FY2024 (to 28 December 2024)","periodEnd":"2024-12-28","source":{"url":"https://www.sec.gov/Archives/edgar/data/2488/000000248826000018/amd-20251227.htm","title":"Advanced Micro Devices Form 10-K for the fiscal year ended December 27, 2025","publisher":"AMD / U.S. SEC","publishedDate":"2026-02-04","accessedDate":"2026-10-22","tier":"T1","locator":"MD&A: net revenue of $12.6 billion in 2024"}}},"baselineRevenueUsd":{"low":16600000000,"central":16600000000,"high":16600000000,"unit":"USD","basis":"Data Center segment revenue, fiscal year, US dollars","period":"FY2025 (to 27 December 2025)","periodEnd":"2025-12-27","source":{"url":"https://www.sec.gov/Archives/edgar/data/2488/000000248826000018/amd-20251227.htm","title":"Advanced Micro Devices Form 10-K for the fiscal year ended December 27, 2025","publisher":"AMD / U.S. SEC","publishedDate":"2026-02-04","accessedDate":"2026-10-22","tier":"T1","locator":"MD&A: Data Center net revenue of $16.6 billion"},"note":"Data Center segment only; it includes EPYC sold to non-AI servers."},"screening":{"merchant-accelerators":{"levelLow":1,"levelHigh":3},"networking-silicon":{"levelLow":1,"levelHigh":2},"host-cpus":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"a105a1c78d12f8eced316ae305fdb4aa25335b99f1abb4502ae4e6623b891166"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"a105a1c78d12f8eced316ae305fdb4aa25335b99f1abb4502ae4e6623b891166"}}},"alphabet":{"schema":"5lc-index-inputs/1","key":"chips-chip-design","entityId":"alphabet","name":"Alphabet","kind":"public","commercialActivityVerified":true,"shares":{"merchant-accelerators":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/1652044/000165204426000018/goog-20251231.htm","title":"Alphabet Inc. Form 10-K for the fiscal year ended December 31, 2025","publisher":"Alphabet / U.S. SEC","publishedDate":"2026-02-05","accessedDate":"2026-10-22","tier":"T1","locator":"Item 1: TPUs are used in Google's own data centres and offered through Google Cloud; Google did not sell accelerators as merchant products in 2025"}},"custom-accelerators":{"low":19.46,"central":22.71,"high":27.58,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.sec.gov/Archives/edgar/data/1652044/000165204426000018/goog-20251231.htm","title":"Alphabet Inc. Form 10-K for the fiscal year ended December 31, 2025","publisher":"Alphabet / U.S. SEC","publishedDate":"2026-02-05","accessedDate":"2026-10-22","tier":"T5","locator":"Item 1: custom-built TPUs, such as Ironwood, our seventh-generation TPU"},"note":"T5 derivation (full inputs). Google TPUs are designed with Broadcom: Google is taken as about 70% (range 60–85%) of Broadcom's roughly $12.5B of fiscal 2025 XPU revenue, so the TPU programme is about $8.75B and Google's owner half $4.375B. Broadcom does not report revenue by customer (Q4 FY2025 call, https://www.fool.com/earnings/call-transcripts/2025/12/12/broadcom-avgo-q4-2025-earnings-call-transcript/). MediaTek's TPU v7e only enters risk production in 2026. 2025 value about $4.375B. Function total $19.27B (was $17.03B before Amazon's programme was re-derived from chip volume, a recorded method decision; see fixes/a recorded method decision/itemisation-changes-a recorded method decision.md) = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $14.7–20.5B (central $16.5B)). Share = 4.375 / 19.27 = 22.71%. Value range $3.75-5.31B gives the share range 19.46-27.58%. Per-unit anchor (a recorded method decision): the $8.75B TPU programme divided by about 2.854 million TPUs shipped in 2025 (GF Securities Brokerage estimate, T4, reported by 24/7 Wall St. on 25 August 2026, https://247wallst.com/investing/2026/08/25/google-tpu-volume-could-triple-to-8-8-million-by-2027/) is about $3,066 per chip; the programme value is unchanged. Itemisation and 2024 prior in a local evidence capture."},"networking-silicon":{"low":0,"central":0.1077,"high":0.4307,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.sec.gov/Archives/edgar/data/1652044/000165204426000018/goog-20251231.htm","title":"Alphabet Inc. Form 10-K for the fiscal year ended December 31, 2025","publisher":"Alphabet / U.S. SEC","publishedDate":"2026-02-05","accessedDate":"2026-10-22","tier":"T5","locator":"Item 1: technical infrastructure"},"note":"T5 derivation (full inputs). Google-designed network chips in AI clusters (TPU interconnect is inside the TPU and counted there); estimated at $0–0.2B. 2025 value about $0.05B. Function total $46.44B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $35.1–47.1B (central $44.8B)). Share = 0.05 / 46.44 = 0.1077%. Value range $0-0.2B gives the share range 0-0.4307%. Itemisation and 2024 prior in a local evidence capture."},"host-cpus":{"low":0.1994,"central":0.9971,"high":2.991,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.sec.gov/Archives/edgar/data/1652044/000165204426000018/goog-20251231.htm","title":"Alphabet Inc. Form 10-K for the fiscal year ended December 31, 2025","publisher":"Alphabet / U.S. SEC","publishedDate":"2026-02-05","accessedDate":"2026-10-22","tier":"T5","locator":"Item 1: technical infrastructure"},"note":"T5 derivation (full inputs). Axion Arm processors used as AI-server hosts, estimated at $0.02–0.3B (card: cloud companies' Arm hosts about $0.5B in total). 2025 value about $0.1B. Function total $10.03B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $7.0–12.9B (central $9.6B)). Share = 0.1 / 10.03 = 0.9971%. Value range $0.02-0.3B gives the share range 0.1994-2.991%. Itemisation and 2024 prior in a local evidence capture."},"processor-ip":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/1652044/000165204426000018/goog-20251231.htm","title":"Alphabet Inc. Form 10-K for the fiscal year ended December 31, 2025","publisher":"Alphabet / U.S. SEC","publishedDate":"2026-02-05","accessedDate":"2026-10-22","tier":"T1","locator":"Item 1: Google does not license processor cores"}}},"screening":{"custom-accelerators":{"levelLow":2,"levelHigh":4},"networking-silicon":{"levelLow":1,"levelHigh":2},"host-cpus":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"10eb0eee4d30fdba205b26957cae13b24517128fa6fd566d2f827516d63d3da4"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"10eb0eee4d30fdba205b26957cae13b24517128fa6fd566d2f827516d63d3da4"}}},"amazon":{"schema":"5lc-index-inputs/1","key":"chips-chip-design","entityId":"amazon","name":"Amazon","kind":"public","commercialActivityVerified":true,"shares":{"merchant-accelerators":{"participates":false,"evidence":{"url":"https://www.nextplatform.com/2025/10/31/aws-bullish-on-homegrown-trainium-ai-accelerators/","title":"AWS 'bullish' on homegrown Trainium AI accelerators","publisher":"The Next Platform (Timothy Prickett Morgan)","publishedDate":"2025-10-31","accessedDate":"2026-10-22","tier":"T2","locator":"Article: Trainium capacity is used inside AWS and rented to customers; Amazon did not sell accelerators as merchant products in 2025"}},"custom-accelerators":{"low":16.15,"central":16.15,"high":16.15,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.aboutamazon.com/news/company-news/amazon-earnings-q4-2025-report","title":"Amazon Q4 2025 earnings report: Read the release","publisher":"Amazon (aboutamazon.com)","publishedDate":"2026-02-05","accessedDate":"2026-10-22","tier":"T5","locator":"Highlights since the last earnings announcement: 'Trainium2 is fully subscribed with 1.4 million chips landed' and 'Project Rainier ... 500,000+ Trainium2 chips'"},"note":"T5 derivation (a recorded method decision: re-derived from chip volume). Programme value = 2025 Trainium chips x a unit value. Volume: Amazon's Q4 2025 release says 'Trainium2 is fully subscribed with 1.4 million chips landed' and that Project Rainier has '500,000+ Trainium2 chips' (the page shows no dateline; its guidance is 'as of February 5, 2026', the release date), and its Q1 2026 release (https://www.aboutamazon.com/news/company-news/amazon-earnings-q1-2026-report) says Amazon 'landed 2.1 million+ AI chips over the past 12 months, more than half of which were Trainium'. The 1.4M is cumulative from Trainium2's December 2024 launch to early February 2026, so about 1.3M is taken as 2025 landings (judgment, range 1.0-1.4M); Trainium1 and Inferentia2 volumes are not reported and are left out. Unit value: the TPU derivation (Alphabet's file) gives a $8.75B programme for about 2.854 million TPUs in 2025 (GF Securities Brokerage estimate, T4, reported by 24/7 Wall St. on 25 August 2026, https://247wallst.com/investing/2026/08/25/google-tpu-volume-could-triple-to-8-8-million-by-2027/), which is $3,066 per chip; Trainium2 is taken at the same value (judgment: a comparable HBM accelerator, range 70-130%). 1.3M x $3,066 = $3.986B of owner's cost for the programme. The design partners' revenue covers only part of it: Marvell about $1.25B of its $1.5B custom revenue (Marvell call of 5 March 2026, https://www.fool.com/earnings/call-transcripts/2026/03/05/marvell-mrvl-q4-2026-earnings-call-transcript/; the Amazon share is a judgment) and Alchip about $0.5B (judgment) make $1.75B, 44% of the programme; Annapurna Labs is Amazon's own chip-design unit (Tom's Hardware, 21 May 2026, https://www.tomshardware.com/tech-industry/semiconductors/custom-ai-asics-examined-from-broadcom-to-mtia); the judgment here is that it designs much of the chip and that Amazon buys much of the wafer, HBM and packaging supply directly, so the other $2.236B is in-house. Split as for Google: Amazon holds the owner half ($1.993B); the partner half ($1.993B) goes to Marvell ($0.625B) and Alchip ($0.25B), each half of its own revenue and below its sourced custom revenue, and the rest ($1.118B) is in-house design and supply, credited to Amazon as for T-Head and Kunlunxin. Amazon's credit is $0.875B + $2.236B = $3.111B. Context: Jassy says Trainium2 is fully subscribed with a multi-billion-dollar run rate (https://www.nextplatform.com/2025/10/31/aws-bullish-on-homegrown-trainium-ai-accelerators/). The earlier $0.875B counted only the partners' half of their revenue. 2025 value about $3.111B. Function total $19.27B (was $17.03B before Amazon's programme was re-derived from chip volume, a recorded method decision; see fixes/a recorded method decision/itemisation-changes-a recorded method decision.md) = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $14.7–20.5B (central $16.5B)). Share = 3.111 / 19.27 = 16.15%. Point value: the engine applies the T5 width. Itemisation and 2024 prior in a local evidence capture."},"networking-silicon":{"low":0.1077,"central":0.4307,"high":1.077,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.nextplatform.com/2025/10/31/aws-bullish-on-homegrown-trainium-ai-accelerators/","title":"AWS 'bullish' on homegrown Trainium AI accelerators","publisher":"The Next Platform (Timothy Prickett Morgan)","publishedDate":"2025-10-31","accessedDate":"2026-10-22","tier":"T5","locator":"AWS chips: Nitro cards and EFA networking"},"note":"T5 derivation (full inputs). Nitro cards and EFA networking in AI servers, at Amazon's cost, estimated at $0.05–0.5B. 2025 value about $0.2B. Function total $46.44B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $35.1–47.1B (central $44.8B)). Share = 0.2 / 46.44 = 0.4307%. Value range $0.05-0.5B gives the share range 0.1077-1.077%. Itemisation and 2024 prior in a local evidence capture."},"host-cpus":{"low":0.4986,"central":1.994,"high":4.986,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.nextplatform.com/2025/10/31/aws-bullish-on-homegrown-trainium-ai-accelerators/","title":"AWS 'bullish' on homegrown Trainium AI accelerators","publisher":"The Next Platform (Timothy Prickett Morgan)","publishedDate":"2025-10-31","accessedDate":"2026-10-22","tier":"T5","locator":"AWS chips: Graviton, Trainium, Nitro"},"note":"T5 derivation (full inputs). Graviton processors used as AI-server hosts, estimated at $0.05–0.5B. 2025 value about $0.2B. Function total $10.03B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $7.0–12.9B (central $9.6B)). Share = 0.2 / 10.03 = 1.994%. Value range $0.05-0.5B gives the share range 0.4986-4.986%. Itemisation and 2024 prior in a local evidence capture."},"processor-ip":{"participates":false,"evidence":{"url":"https://www.nextplatform.com/2025/10/31/aws-bullish-on-homegrown-trainium-ai-accelerators/","title":"AWS 'bullish' on homegrown Trainium AI accelerators","publisher":"The Next Platform (Timothy Prickett Morgan)","publishedDate":"2025-10-31","accessedDate":"2026-10-22","tier":"T2","locator":"Article: AWS designs chips for its own use; no processor-core licensing"}}},"committedExpansion":{"low":11.25,"central":11.25,"high":11.25,"unit":"ratio","basis":"Trainium revenue commitments ÷ annual run rate of the custom-silicon business (Graviton, Trainium, Nitro)","period":"Q1 2026 (call of April 2026)","periodEnd":"2026-03-31","source":{"url":"https://www.aboutamazon.com/news/company-news/amazon-ceo-andy-jassy-amazon-chips-business-q1-2026-earnings","title":"Andy Jassy weighs in on the rapid growth of Amazon's chips business","publisher":"Amazon (aboutamazon.com): excerpt of CEO Andy Jassy's remarks on the Q1 2026 earnings call","publishedDate":"2026-04-29","accessedDate":"2026-10-22","tier":"T2","locator":"Excerpt: 'our annual revenue run rate is now over $20 billion' for the chips business; 'we now have over $225 billion in revenue commitments for Trainium'"},"note":"The company says (CEO Andy Jassy, Q1 2026 earnings call of 29 April 2026, excerpt published by Amazon at https://www.aboutamazon.com/news/company-news/amazon-ceo-andy-jassy-amazon-chips-business-q1-2026-earnings; the page shows no dateline, 29 April 2026 is its page-metadata publication date and the date of the call): 'our annual revenue run rate is now over $20 billion, and growing triple digit percentages year-over-year' and 'we now have over $225 billion in revenue commitments for Trainium'. Amazon's Q1 2026 release (https://www.aboutamazon.com/news/company-news/amazon-earnings-q1-2026-report) says the $20B run rate is for 'Amazon's chips business—inclusive of Graviton, Trainium, and Nitro'. 225 / 20 = 11.25. The numerator is Trainium only and spans several years (the release says the OpenAI commitment of about 2 GW begins ramping in 2027 and Anthropic will secure up to 5 GW), while the denominator is one year of all three chips, and both figures are 'over' amounts, so the ratio is approximate. The Q1 2026 pair is used because Amazon gave no newer commitment figure that I opened."},"screening":{"custom-accelerators":{"levelLow":2,"levelHigh":4},"networking-silicon":{"levelLow":1,"levelHigh":3},"host-cpus":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"d7863ca557bb9f3980dde4db9c0bfc551e82bf3c3d2c9a2b63699c0a625d7249"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"d7863ca557bb9f3980dde4db9c0bfc551e82bf3c3d2c9a2b63699c0a625d7249"}}},"marvell":{"schema":"5lc-index-inputs/1","key":"chips-chip-design","entityId":"marvell","name":"Marvell","kind":"public","commercialActivityVerified":true,"shares":{"merchant-accelerators":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm","title":"Marvell Technology Form 10-K for the fiscal year ended January 31, 2026","publisher":"Marvell / U.S. SEC","publishedDate":"2026-03-11","accessedDate":"2026-10-22","tier":"T1","locator":"Item 1 Business: Marvell designs custom accelerators for their owners and sells interconnect, switching and storage chips; no merchant accelerators"}},"custom-accelerators":{"low":3.893,"central":3.893,"high":3.893,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2026 (to 31 January 2026)","periodEnd":"2026-01-31","source":{"url":"https://www.fool.com/earnings/call-transcripts/2026/03/05/marvell-mrvl-q4-2026-earnings-call-transcript/","title":"Marvell (MRVL) Q4 2026 Earnings Call Transcript","publisher":"The Motley Fool (transcript of Marvell's earnings call of 5 March 2026)","publishedDate":"2026-03-05","accessedDate":"2026-10-22","tier":"T5","locator":"Prepared remarks, custom business: 'we have scaled from zero revenue to $1.5 billion in fiscal 2026 ... we doubled our custom revenue year over year'"},"note":"T5 derivation (fix pass). Marvell management says on the Q4 FY2026 call (5 March 2026; transcript https://www.fool.com/earnings/call-transcripts/2026/03/05/marvell-mrvl-q4-2026-earnings-call-transcript/): 'In just a few years, we have scaled from zero revenue to $1.5 billion in fiscal 2026. As you may recall, the first meaningful ramp began in fiscal 2025. Fiscal 2026 marked the first full year of production for those programs, and as a result, we doubled our custom revenue year over year.' The 10-K (https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm) does not report custom revenue; it reports data center revenue of $6,100.3M. The $1.5B is taken as AI accelerator and attach programme value (it may include some custom work for non-AI uses: Item 1 says custom ASICs serve AI, data center, compute, networking, carrier, storage and aerospace and defense), and Marvell's design-partner half is $0.75B. Marvell does not report customers; programmes are the author's judgment (Amazon Trainium about $1.25B, Microsoft about $0.15B, others about $0.1B). 2025 value about $0.75B. Function total $19.27B (was $17.03B before Amazon's programme was re-derived from chip volume, a recorded method decision; see fixes/a recorded method decision/itemisation-changes-a recorded method decision.md) = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $14.7–20.5B (central $16.5B)). Share = 0.75 / 19.27 = 3.893%. Point value: the engine applies the T5 width. Itemisation and 2024 prior in a local evidence capture (adjustments made in the fix pass: full-inputs/chips-chip-design/fixes/itemisation-changes.md)."},"networking-silicon":{"low":6.46,"central":6.46,"high":6.46,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2026 (to 31 January 2026)","periodEnd":"2026-01-31","source":{"url":"https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm","title":"Marvell Technology Form 10-K for the fiscal year ended January 31, 2026","publisher":"Marvell / U.S. SEC","publishedDate":"2026-03-11","accessedDate":"2026-10-22","tier":"T5","locator":"MD&A: Data center $6,100.3M"},"note":"T5 derivation (fix pass). Electro-optics DSPs, switching and interconnect: fiscal 2026 data-centre revenue of $6,100.3M (10-K, https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm) less custom silicon of about $1.5B (Marvell management, Q4 FY2026 call, https://www.fool.com/earnings/call-transcripts/2026/03/05/marvell-mrvl-q4-2026-earnings-call-transcript/) leaves about $4.6B; storage controllers and other data-centre products are taken as about $1.6B, which is the author's judgment (no opened source gives Marvell's storage revenue), so interconnect and switching are about $3.0B. 2025 value about $3B. Function total $46.44B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $35.1–47.1B (central $44.8B)). Share = 3 / 46.44 = 6.46%. Point value: the engine applies the T5 width. Itemisation and 2024 prior in a local evidence capture (adjustments made in the fix pass: full-inputs/chips-chip-design/fixes/itemisation-changes.md)."},"host-cpus":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm","title":"Marvell Technology Form 10-K for the fiscal year ended January 31, 2026","publisher":"Marvell / U.S. SEC","publishedDate":"2026-03-11","accessedDate":"2026-10-22","tier":"T1","locator":"Item 1 Business: no merchant server processors"}},"processor-ip":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm","title":"Marvell Technology Form 10-K for the fiscal year ended January 31, 2026","publisher":"Marvell / U.S. SEC","publishedDate":"2026-03-11","accessedDate":"2026-10-22","tier":"T1","locator":"Item 1 Business: no processor-core licensing"}}},"priorShares":{"custom-accelerators":{"low":2.847,"central":2.847,"high":2.847,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2025 (to 1 February 2025)","periodEnd":"2025-02-01","source":{"url":"https://www.fool.com/earnings/call-transcripts/2026/03/05/marvell-mrvl-q4-2026-earnings-call-transcript/","title":"Marvell (MRVL) Q4 2026 Earnings Call Transcript","publisher":"The Motley Fool (transcript of Marvell's earnings call of 5 March 2026)","publishedDate":"2026-03-05","accessedDate":"2026-10-22","tier":"T5","locator":"Prepared remarks, custom business: first meaningful ramp in fiscal 2025; 'we doubled our custom revenue year over year' to $1.5 billion in fiscal 2026"},"note":"T5 derivation of the prior-year share. Marvell management says on the Q4 FY2026 call (5 March 2026; transcript https://www.fool.com/earnings/call-transcripts/2026/03/05/marvell-mrvl-q4-2026-earnings-call-transcript/): 'In just a few years, we have scaled from zero revenue to $1.5 billion in fiscal 2026. As you may recall, the first meaningful ramp began in fiscal 2025. Fiscal 2026 marked the first full year of production for those programs, and as a result, we doubled our custom revenue year over year.' The first sentence's 'zero' is the start of the ramp a few years ago; it is not the fiscal 2025 base. Fiscal 2025 custom revenue is therefore about half of $1.5B, $0.75B (the company says 'doubled', so the base is approximate). Marvell's design-partner half is $0.375B. 2024 value about $0.375B. Function total $13.17B (was $11.21B before Amazon's 2024 programme was restated with the same in-house multiple, a recorded method decision) = the itemised 2024 values of every candidate plus residual others, built like the 2025 total; candidates without a reported 2024 value are set at half their 2025 value (an assumption: most grew at least 100%). Share = 0.375 / 13.174 = 2.847%. Point value: the engine applies the T5 width. Itemisation and 2024 prior in a local evidence capture (adjustments made in the fix pass: full-inputs/chips-chip-design/fixes/itemisation-changes.md)."},"networking-silicon":{"low":7.035,"central":9.849,"high":12.19,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2025 (to 1 February 2025)","periodEnd":"2025-02-01","source":{"url":"https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm","title":"Marvell Technology Form 10-K for the fiscal year ended January 31, 2026","publisher":"Marvell / U.S. SEC","publishedDate":"2026-03-11","accessedDate":"2026-10-22","tier":"T5","locator":"MD&A: Data center $4,164.2M in fiscal 2025"},"note":"T5 derivation of the prior-year share. Electro-optics DSPs, switching and interconnect: data-centre revenue $4.164B (10-K) less custom silicon of about $0.75B (half of the $1.5B the company reports for fiscal 2026, because it says custom doubled) and storage and other of about $1.3B (the author's judgment; no opened source gives it) leaves about $2.1B (the range $1.5-2.6B reflects the storage judgment). 2024 value about $2.1B. Function total $21.32B = the itemised 2024 values of every candidate plus residual others, built like the 2025 total; candidates without a reported 2024 value are set at half their 2025 value (an assumption: most grew at least 100%). Share = 2.1 / 21.32 = 9.849%. Value range $1.5-2.6B gives the share range 7.035-12.19%. Itemisation and 2024 prior in a local evidence capture (adjustments made in the fix pass: full-inputs/chips-chip-design/fixes/itemisation-changes.md)."}},"activityGrowth":{"current":{"low":6100300000,"central":6100300000,"high":6100300000,"unit":"USD","basis":"Data center revenue, fiscal year, US dollars","period":"FY2026 (to 31 January 2026)","periodEnd":"2026-01-31","source":{"url":"https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm","title":"Marvell Technology Form 10-K for the fiscal year ended January 31, 2026","publisher":"Marvell / U.S. SEC","publishedDate":"2026-03-11","accessedDate":"2026-10-22","tier":"T1","locator":"MD&A: Data center $6,100.3M"},"note":"Data center revenue (custom ASICs, electro-optics, switching, storage) is the closest reported measure."},"prior":{"low":4164200000,"central":4164200000,"high":4164200000,"unit":"USD","basis":"Data center revenue, fiscal year, US dollars","period":"FY2025 (to 1 February 2025)","periodEnd":"2025-02-01","source":{"url":"https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm","title":"Marvell Technology Form 10-K for the fiscal year ended January 31, 2026","publisher":"Marvell / U.S. SEC","publishedDate":"2026-03-11","accessedDate":"2026-10-22","tier":"T1","locator":"MD&A: Data center $4,164.2M"}}},"baselineRevenueUsd":{"low":6100300000,"central":6100300000,"high":6100300000,"unit":"USD","basis":"Data center revenue, fiscal year, US dollars","period":"FY2026 (to 31 January 2026)","periodEnd":"2026-01-31","source":{"url":"https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm","title":"Marvell Technology Form 10-K for the fiscal year ended January 31, 2026","publisher":"Marvell / U.S. SEC","publishedDate":"2026-03-11","accessedDate":"2026-10-22","tier":"T1","locator":"MD&A: Data center $6,100.3M"},"note":"Data center end market only, never Marvell's total ($8,194.6M)."},"screening":{"custom-accelerators":{"levelLow":2,"levelHigh":4},"networking-silicon":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"9db2bf0cf61423397c9460284f1ee5bc29c2957441522cf7142742713d4685ca"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"9db2bf0cf61423397c9460284f1ee5bc29c2957441522cf7142742713d4685ca"}}},"intel":{"schema":"5lc-index-inputs/1","key":"chips-chip-design","entityId":"intel","name":"Intel","kind":"public","commercialActivityVerified":true,"shares":{"merchant-accelerators":{"low":0.0569,"central":0.1707,"high":0.2845,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2025 (to 27 December 2025)","periodEnd":"2025-12-27","source":{"url":"https://www.sec.gov/Archives/edgar/data/50863/000005086326000011/intc-20251227.htm","title":"Intel Corporation Form 10-K for the fiscal year ended December 27, 2025","publisher":"Intel / U.S. SEC","publishedDate":"2026-01-23","accessedDate":"2026-10-22","tier":"T5","locator":"MD&A: lower 2025 Gaudi AI accelerator inventory-related charges"},"note":"T5 derivation (full inputs). Gaudi accelerators: Intel reports no Gaudi revenue and says 2025 Gaudi inventory charges were lower than in 2024; estimated at $0.1–0.5B. 2025 value about $0.3B. Function total $175.74B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $168.7–183.3B (central $175.8B)). Share = 0.3 / 175.74 = 0.1707%. Value range $0.1-0.5B gives the share range 0.0569-0.2845%. Itemisation and 2024 prior in a local evidence capture."},"custom-accelerators":{"low":0.9473,"central":0.9473,"high":0.9473,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2025 (to 27 December 2025)","periodEnd":"2025-12-27","source":{"url":"https://d1io3yog0oux5.cloudfront.net/_3ccf9cf30376bf77bbb27f582e51d00d/intel/db/887/9159/prepared_remarks/Intel-4Q2025-Earnings-Call+1+(1).pdf","title":"CEO/CFO Earnings Call Comments: prepared remarks on Intel's fourth-quarter 2025 earnings call (22 January 2026)","publisher":"Intel","publishedDate":"2026-01-22","accessedDate":"2026-10-22","tier":"T5","locator":"CFO section, PDF page 5: 'our custom ASIC business grew more than 50% in 2025, 26% sequentially, and reached an annualized revenue run rate greater than $1 billion in Q4'"},"note":"T5 derivation (fix pass). The 10-K (https://www.sec.gov/Archives/edgar/data/50863/000005086326000011/intc-20251227.htm, Item 1, DCAI) lists custom ASICs among DCAI's products and says 'we also compete with Broadcom in the custom ASICs development market', so participation is recorded, not denied; the 10-K reports no ASIC revenue. The company says in its Q4 2025 prepared remarks that 'due to the networking demand for the AI infrastructure buildout, our custom ASIC business grew more than 50% in 2025, 26% sequentially, and reached an annualized revenue run rate greater than $1 billion in Q4'. Derivation (judgments): Q4 annualised rate taken as $1.0B, so Q4 is $0.25B; if each earlier quarter was 1/1.26 of the next, 2025 is 0.25 + 0.198 + 0.157 + 0.125 = about $0.73B; about half of that taken as AI-related programme value (range 25-100%: the company ties the growth to AI infrastructure demand but does not split the base from older telecom ASIC work), $0.365B; Intel's design-partner half is $0.1825B. This value is counted here once and not in networking-silicon, whose Intel estimate covers Ethernet adapters and IPUs only; part of it may be networking rather than accelerator value, which the engine's T5 width covers. 2025 value about $0.1825B. Function total $19.27B (was $17.03B before Amazon's programme was re-derived from chip volume, a recorded method decision; see fixes/a recorded method decision/itemisation-changes-a recorded method decision.md) = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $14.7–20.5B (central $16.5B)). Share = 0.1825 / 19.27 = 0.9473%. Point value: the engine applies the T5 width. Itemisation and 2024 prior in a local evidence capture (adjustments made in the fix pass: full-inputs/chips-chip-design/fixes/itemisation-changes.md)."},"networking-silicon":{"low":0.2153,"central":0.5384,"high":1.077,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2025 (to 27 December 2025)","periodEnd":"2025-12-27","source":{"url":"https://www.sec.gov/Archives/edgar/data/50863/000005086326000011/intc-20251227.htm","title":"Intel Corporation Form 10-K for the fiscal year ended December 27, 2025","publisher":"Intel / U.S. SEC","publishedDate":"2026-01-23","accessedDate":"2026-10-22","tier":"T5","locator":"Item 1: Ethernet controllers and IPUs in DCAI"},"note":"T5 derivation (full inputs). Ethernet adapters and IPUs used in AI clusters; not reported separately, estimated at $0.1–0.5B. 2025 value about $0.25B. Function total $46.44B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $35.1–47.1B (central $44.8B)). Share = 0.25 / 46.44 = 0.5384%. Value range $0.1-0.5B gives the share range 0.2153-1.077%. Itemisation and 2024 prior in a local evidence capture."},"host-cpus":{"low":27.92,"central":27.92,"high":27.92,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2025 (to 27 December 2025)","periodEnd":"2025-12-27","source":{"url":"https://www.sec.gov/Archives/edgar/data/50863/000005086326000011/intc-20251227.htm","title":"Intel Corporation Form 10-K for the fiscal year ended December 27, 2025","publisher":"Intel / U.S. SEC","publishedDate":"2026-01-23","accessedDate":"2026-10-22","tier":"T5","locator":"MD&A: DCAI revenue $16,919M (2025) and $16,125M (2024); 2025 server volume +9%, server ASPs -4%; other DCAI product revenue $2.7B in 2024"},"note":"T5 derivation (fix pass). The 10-K gives other DCAI product revenue of $2.7B for 2024, so 2024 server revenue is about $13.425B ($16.125B less $2.7B). For 2025 it gives no figure for other DCAI products (it says only that they increased, on networking demand), but it says server volume rose 9% and server ASPs fell 4%: $13.425B x 1.09 x 0.96 = about $14.05B of Xeon server revenue, which implies other DCAI products of about $2.87B (derived, not reported). 20% of server revenue (about $2.81B, rounded to $2.8B) taken as AI-server hosts (an assumption). 2025 value about $2.8B. Function total $10.03B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $7.0–12.9B (central $9.6B)). Share = 2.8 / 10.03 = 27.92%. Point value: the engine applies the T5 width. Itemisation and 2024 prior in a local evidence capture (adjustments made in the fix pass: full-inputs/chips-chip-design/fixes/itemisation-changes.md)."},"processor-ip":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/50863/000005086326000011/intc-20251227.htm","title":"Intel Corporation Form 10-K for the fiscal year ended December 27, 2025","publisher":"Intel / U.S. SEC","publishedDate":"2026-01-23","accessedDate":"2026-10-22","tier":"T1","locator":"Item 1 Business: Intel does not license its x86 processor cores to other chip designers"}}},"priorShares":{"host-cpus":{"low":48.75,"central":48.75,"high":48.75,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2024 (to 28 December 2024)","periodEnd":"2024-12-28","source":{"url":"https://www.sec.gov/Archives/edgar/data/50863/000005086326000011/intc-20251227.htm","title":"Intel Corporation Form 10-K for the fiscal year ended December 27, 2025","publisher":"Intel / U.S. SEC","publishedDate":"2026-01-23","accessedDate":"2026-10-22","tier":"T5","locator":"MD&A: DCAI revenue $16,125M in 2024; other DCAI product revenue $2.7B in 2024"},"note":"T5 derivation of the prior-year share. Xeon server processors estimated at $13.4B (DCAI $16.125B less $2.7B of other DCAI products); 20% ($2.69B) taken as AI-server hosts, the same method as 2025. 2024 value about $2.685B. Function total $5.51B = the itemised 2024 values of every candidate plus residual others, built like the 2025 total; candidates without a reported 2024 value are set at half their 2025 value (an assumption: most grew at least 100%). Share = 2.685 / 5.51 = 48.75%. Point value: the engine applies the T5 width. Itemisation in a local evidence capture."},"merchant-accelerators":{"low":0.134,"central":0.2679,"high":0.4465,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2024 (to 28 December 2024)","periodEnd":"2024-12-28","source":{"url":"https://www.constellationr.com/insights/news/intel-defends-gaudi-3-it-misses-2024-sales-targets","title":"Intel defends Gaudi 3 as it misses 2024 sales targets","publisher":"Constellation Research (Larry Dignan)","publishedDate":"2024-11-01","accessedDate":"2026-10-22","tier":"T5","locator":"CEO Gelsinger, Q3 2024 call: we will not achieve our target of $500 million in revenue for Gaudi in 2024"},"note":"T5 derivation of the prior-year share. Intel said in October 2024 that it would miss its $500M Gaudi revenue target for 2024; no figure was reported, so Gaudi is estimated at $0.15-0.5B (central $0.3B). 2024 value about $0.3B. Function total $111.97B = the itemised 2024 values of every candidate plus residual others, built like the 2025 total; candidates without a reported 2024 value are set at half their 2025 value (an assumption: most grew at least 100%). Share = 0.3 / 111.97 = 0.2679%. Value range $0.15-0.5B gives the share range 0.134-0.4465%. Itemisation in a local evidence capture."},"networking-silicon":{"low":0.469,"central":1.173,"high":2.345,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2024 (to 28 December 2024)","periodEnd":"2024-12-28","source":{"url":"https://www.sec.gov/Archives/edgar/data/50863/000005086326000011/intc-20251227.htm","title":"Intel Corporation Form 10-K for the fiscal year ended December 27, 2025","publisher":"Intel / U.S. SEC","publishedDate":"2026-01-23","accessedDate":"2026-10-22","tier":"T5","locator":"Item 1: Ethernet controllers and IPUs in DCAI"},"note":"T5 derivation of the prior-year share. Ethernet adapters and IPUs used in AI clusters; not reported separately, estimated at $0.1-0.5B (central $0.25B). 2024 value about $0.25B. Function total $21.32B = the itemised 2024 values of every candidate plus residual others, built like the 2025 total; candidates without a reported 2024 value are set at half their 2025 value (an assumption: most grew at least 100%). Share = 0.25 / 21.32 = 1.173%. Value range $0.1-0.5B gives the share range 0.469-2.345%. Itemisation in a local evidence capture."},"custom-accelerators":{"low":0.9223,"central":0.9223,"high":0.9223,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2024 (to 28 December 2024)","periodEnd":"2024-12-28","source":{"url":"https://d1io3yog0oux5.cloudfront.net/_3ccf9cf30376bf77bbb27f582e51d00d/intel/db/887/9159/prepared_remarks/Intel-4Q2025-Earnings-Call+1+(1).pdf","title":"CEO/CFO Earnings Call Comments: prepared remarks on Intel's fourth-quarter 2025 earnings call (22 January 2026)","publisher":"Intel","publishedDate":"2026-01-22","accessedDate":"2026-10-22","tier":"T5","locator":"CFO section, PDF page 5: 'our custom ASIC business grew more than 50% in 2025, 26% sequentially, and reached an annualized revenue run rate greater than $1 billion in Q4'"},"note":"T5 derivation of the prior-year share. The company says its custom ASIC business grew more than 50% in 2025 (Q4 2025 prepared remarks), so 2024 is at most the 2025 AI-related programme value divided by 1.5: $0.365B / 1.5 = $0.243B (an upper bound, because 'more than 50%' is a floor on growth); Intel's half is $0.1215B, the same method as 2025. 2024 value about $0.1215B. Function total $13.17B (was $11.21B before Amazon's 2024 programme was restated with the same in-house multiple, a recorded method decision) = the itemised 2024 values of every candidate plus residual others, built like the 2025 total; candidates without a reported 2024 value are set at half their 2025 value (an assumption: most grew at least 100%). Share = 0.1215 / 13.174 = 0.9223%. Point value: the engine applies the T5 width. Itemisation and 2024 prior in a local evidence capture (adjustments made in the fix pass: full-inputs/chips-chip-design/fixes/itemisation-changes.md)."}},"activityGrowth":{"current":{"low":16919000000,"central":16919000000,"high":16919000000,"unit":"USD","basis":"DCAI segment revenue, fiscal year, US dollars","period":"FY2025 (to 27 December 2025)","periodEnd":"2025-12-27","source":{"url":"https://www.sec.gov/Archives/edgar/data/50863/000005086326000011/intc-20251227.htm","title":"Intel Corporation Form 10-K for the fiscal year ended December 27, 2025","publisher":"Intel / U.S. SEC","publishedDate":"2026-01-23","accessedDate":"2026-10-22","tier":"T1","locator":"MD&A, Intel Products Financial Performance: DCAI revenue $16,919M"},"note":"DCAI (Xeon, Gaudi, NICs, IPUs, custom ASICs) is the closest reported measure; most of it is server processors sold to non-AI servers too."},"prior":{"low":16125000000,"central":16125000000,"high":16125000000,"unit":"USD","basis":"DCAI segment revenue, fiscal year, US dollars","period":"FY2024 (to 28 December 2024)","periodEnd":"2024-12-28","source":{"url":"https://www.sec.gov/Archives/edgar/data/50863/000005086326000011/intc-20251227.htm","title":"Intel Corporation Form 10-K for the fiscal year ended December 27, 2025","publisher":"Intel / U.S. SEC","publishedDate":"2026-01-23","accessedDate":"2026-10-22","tier":"T1","locator":"MD&A, Intel Products Financial Performance: DCAI revenue $16,125M (2024)"}}},"baselineRevenueUsd":{"low":16919000000,"central":16919000000,"high":16919000000,"unit":"USD","basis":"DCAI segment revenue, fiscal year, US dollars","period":"FY2025 (to 27 December 2025)","periodEnd":"2025-12-27","source":{"url":"https://www.sec.gov/Archives/edgar/data/50863/000005086326000011/intc-20251227.htm","title":"Intel Corporation Form 10-K for the fiscal year ended December 27, 2025","publisher":"Intel / U.S. SEC","publishedDate":"2026-01-23","accessedDate":"2026-10-22","tier":"T1","locator":"MD&A, Intel Products Financial Performance: DCAI revenue $16,919M"},"note":"DCAI segment only, never Intel's consolidated total."},"screening":{"merchant-accelerators":{"levelLow":1,"levelHigh":2},"networking-silicon":{"levelLow":1,"levelHigh":2},"host-cpus":{"levelLow":1,"levelHigh":3},"custom-accelerators":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"746764c5a7e31b20d616e535244e70b8b7a3455f0fbe59aae4dab1f183d45c6a"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"746764c5a7e31b20d616e535244e70b8b7a3455f0fbe59aae4dab1f183d45c6a"}}},"arm":{"schema":"5lc-index-inputs/1","key":"chips-chip-design","entityId":"arm","name":"Arm (SoftBank group: Arm, Ampere, Graphcore)","kind":"public","parentId":"softbank-group","commercialActivityVerified":true,"shares":{"merchant-accelerators":{"low":0,"central":0.002845,"high":0.01138,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.verdict.co.uk/softbank-acquires-chip-company-graphcore/","title":"SoftBank acquires chip company Graphcore","publisher":"Verdict","publishedDate":"2024-07-12","accessedDate":"2026-10-22","tier":"T5","locator":"Graphcore is a wholly owned SoftBank subsidiary; 2022 sales $2.7M"},"note":"T5 derivation (full inputs). Graphcore (SoftBank group since July 2024, counted in this entry) reported 2022 sales of $2.7M; its 2025 IPU sales are estimated at $0–0.02B. 2025 value about $0.005B. Function total $175.74B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $168.7–183.3B (central $175.8B)). Share = 0.005 / 175.74 = 0.002845%. Value range $0-0.02B gives the share range 0-0.01138%. Itemisation and 2024 prior in a local evidence capture."},"custom-accelerators":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/1973239/000197323926000097/arm-20260331.htm","title":"Arm Holdings plc Form 20-F for the fiscal year ended March 31, 2026","publisher":"Arm Holdings / U.S. SEC","publishedDate":"2026-05-26","accessedDate":"2026-10-22","tier":"T1","locator":"Item 4/5: Arm licenses IP and sells compute subsystems; it designed no custom accelerator in production for another owner in fiscal 2026"}},"networking-silicon":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/1973239/000197323926000097/arm-20260331.htm","title":"Arm Holdings plc Form 20-F for the fiscal year ended March 31, 2026","publisher":"Arm Holdings / U.S. SEC","publishedDate":"2026-05-26","accessedDate":"2026-10-22","tier":"T1","locator":"Item 4: Arm's products are processor IP and compute subsystems; no networking chips"}},"host-cpus":{"low":0.09971,"central":0.4986,"high":1.994,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://group.softbank/en/news/press/20251126","title":"Completion of acquisition of Ampere Computing Holdings LLC","publisher":"SoftBank Group Corp.","publishedDate":"2025-11-26","accessedDate":"2026-10-22","tier":"T5","locator":"Ampere became a wholly owned SoftBank subsidiary on 25 November 2025"},"note":"T5 derivation (full inputs). Ampere (SoftBank group, counted in this entry) Arm server processors used as AI hosts, estimated at $0.01–0.2B. 2025 value about $0.05B. Function total $10.03B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $7.0–12.9B (central $9.6B)). Share = 0.05 / 10.03 = 0.4986%. Value range $0.01-0.2B gives the share range 0.09971-1.994%. Itemisation and 2024 prior in a local evidence capture."},"processor-ip":{"low":91.91,"central":91.91,"high":91.91,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2026 (to 31 March 2026)","periodEnd":"2026-03-31","source":{"url":"https://www.sec.gov/Archives/edgar/data/1973239/000197323926000097/arm-20260331.htm","title":"Arm Holdings plc Form 20-F for the fiscal year ended March 31, 2026","publisher":"Arm Holdings / U.S. SEC","publishedDate":"2026-05-26","accessedDate":"2026-10-22","tier":"T5","locator":"Item 5.A: royalty revenue $2,613M; licence and other $2,307M"},"note":"T5 derivation (full inputs). Arm data-centre royalties taken as about 12% of fiscal 2026 royalty revenue of $2,613M ($0.31B) plus about 10% of licence and other revenue of $2,307M ($0.23B): $0.545B. Arm does not report data-centre revenue. 2025 value about $0.545B. Function total $0.59B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $0.40–1.0B (central $0.59B)). Share = 0.545 / 0.59 = 91.91%. Point value: the engine applies the T5 width. Itemisation and 2024 prior in a local evidence capture."}},"activityGrowth":{"current":{"low":2,"central":2,"high":2,"unit":"index","basis":"Data-centre royalty revenue, indexed to fiscal 2025 = 1","period":"FY2026 (to 31 March 2026)","periodEnd":"2026-03-31","source":{"url":"https://www.fool.com/earnings/call-transcripts/2026/05/07/arm-arm-q4-2026-earnings-call-transcript","title":"Arm (ARM) Q4 2026 earnings call transcript","publisher":"The Motley Fool (transcript of Arm's earnings call of 6 May 2026)","publishedDate":"2026-05-07","accessedDate":"2026-10-22","tier":"T2","locator":"Management: data center royalty revenue has more than doubled year over year and is expected to double again in fiscal 2027"},"note":"Arm reports no data-centre royalty dollars, so this is an index, not a revenue figure. The company says (Q4 FY2026 call, 6 May 2026) that data-centre royalty revenue more than doubled year over year; 2.0 is therefore a lower bound on the current value. The transcript does not say whether the statement covers the quarter or the fiscal year; it is treated as the fiscal year."},"prior":{"low":1,"central":1,"high":1,"unit":"index","basis":"Data-centre royalty revenue, indexed to fiscal 2025 = 1","period":"FY2025 (to 31 March 2025)","periodEnd":"2025-03-31","source":{"url":"https://www.fool.com/earnings/call-transcripts/2026/05/07/arm-arm-q4-2026-earnings-call-transcript","title":"Arm (ARM) Q4 2026 earnings call transcript","publisher":"The Motley Fool (transcript of Arm's earnings call of 6 May 2026)","publishedDate":"2026-05-07","accessedDate":"2026-10-22","tier":"T2","locator":"Base of the index: fiscal 2025 data-centre royalty revenue"}}},"committedExpansion":{"low":0.421,"central":0.421,"high":0.421,"unit":"ratio","basis":"Remaining performance obligations ÷ fiscal 2026 total revenue","period":"31 March 2026","periodEnd":"2026-03-31","source":{"url":"https://www.sec.gov/Archives/edgar/data/1973239/000197323926000097/arm-20260331.htm","title":"Arm Holdings plc Form 20-F for the fiscal year ended March 31, 2026","publisher":"Arm Holdings / U.S. SEC","publishedDate":"2026-05-26","accessedDate":"2026-10-22","tier":"T1","locator":"Notes to the consolidated financial statements, Remaining Performance Obligations: $2,071.4 million at 31 March 2026"},"note":"RPO $2,071.4M at 31 March 2026 (Note 4; excludes future royalties) ÷ total revenue $4,920M = 0.4210. Company-wide: Arm reports no data-centre split of its RPO."},"screening":{"merchant-accelerators":{"levelLow":1,"levelHigh":2},"host-cpus":{"levelLow":1,"levelHigh":2},"processor-ip":{"levelLow":3,"levelHigh":5}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"2592f2c4c61b871bb2814fbe81d3e8142155a032866c93146b489f22abb52335"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"2592f2c4c61b871bb2814fbe81d3e8142155a032866c93146b489f22abb52335"}}},"alibaba":{"schema":"5lc-index-inputs/1","key":"chips-chip-design","entityId":"alibaba","name":"Alibaba","kind":"public","commercialActivityVerified":true,"shares":{"merchant-accelerators":{"low":0.02731,"central":0.08877,"high":0.2845,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://kfgo.com/?p=1285523","title":"Chinese chipmakers claim nearly half of local market as Nvidia's lead shrinks (IDC data)","publisher":"Reuters (Che Pan and Laurie Chen), via KFGO","publishedDate":"2026-04-01","accessedDate":"2026-10-22","tier":"T5","locator":"IDC 2025: Alibaba's T-Head about 265,000 cards"},"note":"T5 derivation (full inputs). T-Head PPU/Zhenwu cards sold to outside customers. IDC counts about 265,000 T-Head cards in China in 2025 (Reuters, https://kfgo.com/?p=1285523), worth $0.8-2.0B at $3,000-7,500 a card (central $1.3B). Outside sales are a small part: China Unicom's Xining centre alone took 16,384 PPUs, about 6% of the cards (https://www.eetasia.com/?p=69997). Outside sales are taken as 12% (range 6-25%) of the total; the rest is Alibaba's own use, counted under custom accelerators, so nothing is counted twice. 2025 value about $0.156B. Function total $175.74B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $168.7–183.3B (central $175.8B)). Share = 0.156 / 175.74 = 0.08877%. Value range $0.048-0.5B gives the share range 0.02731-0.2845%. Itemisation and 2024 prior in a local evidence capture."},"custom-accelerators":{"low":3.114,"central":5.938,"high":9.759,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.sec.gov/Archives/edgar/data/1577552/000119312526231755/baba-20260331.htm","title":"Alibaba Group Holding Form 20-F for the fiscal year ended March 31, 2026","publisher":"Alibaba Group / U.S. SEC","publishedDate":"2026-05-20","accessedDate":"2026-10-22","tier":"T5","locator":"Item 4: T-Head AI chips in production at scale for Alibaba Cloud"},"note":"T5 derivation (full inputs). T-Head accelerators that Alibaba Cloud uses itself: 88% (range 75-94%) of the $1.3B (range $0.8-2.0B) value of IDC's 265,000 cards in China in 2025 (https://kfgo.com/?p=1285523), valued at the owner's cost as the card does for custom chips. T-Head is Alibaba's wholly owned chip-design unit, so the owner and the design partner are the same group and the whole programme value is credited to Alibaba. This replaces the screening's zero, which counted these cards under merchant accelerators. 2025 value about $1.144B. Function total $19.27B (was $17.03B before Amazon's programme was re-derived from chip volume, a recorded method decision; see fixes/a recorded method decision/itemisation-changes-a recorded method decision.md) = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $14.7–20.5B (central $16.5B)). Share = 1.144 / 19.27 = 5.938%. Value range $0.6-1.88B gives the share range 3.114-9.759%. Itemisation and 2024 prior in a local evidence capture."},"networking-silicon":{"low":0,"central":0.0646,"high":0.2153,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.sec.gov/Archives/edgar/data/1577552/000119312526231755/baba-20260331.htm","title":"Alibaba Group Holding Form 20-F for the fiscal year ended March 31, 2026","publisher":"Alibaba Group / U.S. SEC","publishedDate":"2026-05-20","accessedDate":"2026-10-22","tier":"T5","locator":"Item 4: proprietary chips in cloud infrastructure"},"note":"T5 derivation (full inputs). CIPU infrastructure processors in Alibaba Cloud AI servers, estimated at $0–0.1B. 2025 value about $0.03B. Function total $46.44B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $35.1–47.1B (central $44.8B)). Share = 0.03 / 46.44 = 0.0646%. Value range $0-0.1B gives the share range 0-0.2153%. Itemisation and 2024 prior in a local evidence capture."},"host-cpus":{"low":0.09971,"central":0.4986,"high":1.994,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.sec.gov/Archives/edgar/data/1577552/000119312526231755/baba-20260331.htm","title":"Alibaba Group Holding Form 20-F for the fiscal year ended March 31, 2026","publisher":"Alibaba Group / U.S. SEC","publishedDate":"2026-05-20","accessedDate":"2026-10-22","tier":"T5","locator":"Item 4: T-Head, our chip design subsidiary"},"note":"T5 derivation (full inputs). Yitian Arm processors used as AI-server hosts in Alibaba Cloud, estimated at $0.01–0.2B. 2025 value about $0.05B. Function total $10.03B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $7.0–12.9B (central $9.6B)). Share = 0.05 / 10.03 = 0.4986%. Value range $0.01-0.2B gives the share range 0.09971-1.994%. Itemisation and 2024 prior in a local evidence capture."},"processor-ip":{"low":0,"central":0.8432,"high":3.373,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.sec.gov/Archives/edgar/data/1577552/000119312526231755/baba-20260331.htm","title":"Alibaba Group Holding Form 20-F for the fiscal year ended March 31, 2026","publisher":"Alibaba Group / U.S. SEC","publishedDate":"2026-05-20","accessedDate":"2026-10-22","tier":"T5","locator":"Item 4: T-Head, our chip design subsidiary"},"note":"T5 derivation (full inputs). T-Head XuanTie RISC-V core licences; data-centre part estimated at $0–0.02B. 2025 value about $0.005B. Function total $0.59B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $0.40–1.0B (central $0.59B)). Share = 0.005 / 0.59 = 0.8432%. Value range $0-0.02B gives the share range 0-3.373%. Itemisation and 2024 prior in a local evidence capture."}},"screening":{"merchant-accelerators":{"levelLow":1,"levelHigh":3},"networking-silicon":{"levelLow":1,"levelHigh":2},"host-cpus":{"levelLow":1,"levelHigh":2},"processor-ip":{"levelLow":1,"levelHigh":2},"custom-accelerators":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"28534e05f8cdce8cf0b6a8b22b8b61d78525ea744fabc9382bc0c134a1c669cf"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"28534e05f8cdce8cf0b6a8b22b8b61d78525ea744fabc9382bc0c134a1c669cf"}}},"credo":{"schema":"5lc-index-inputs/1","key":"chips-chip-design","entityId":"credo","name":"Credo","kind":"public","commercialActivityVerified":true,"shares":{"merchant-accelerators":{"participates":false,"evidence":{"url":"https://www.01net.it/credo-technology-group-holding-ltd-reports-fourth-quarter-and-fiscal-year-2026-financial-results/","title":"Credo Technology Group Holding Ltd reports fourth quarter and fiscal year 2026 financial results","publisher":"Credo (press release, via 01net)","publishedDate":"2026-06-02","accessedDate":"2026-10-22","tier":"T2","locator":"Release: Credo sells copper and optical interconnect products, retimers and DSPs only"}},"custom-accelerators":{"participates":false,"evidence":{"url":"https://www.01net.it/credo-technology-group-holding-ltd-reports-fourth-quarter-and-fiscal-year-2026-financial-results/","title":"Credo Technology Group Holding Ltd reports fourth quarter and fiscal year 2026 financial results","publisher":"Credo (press release, via 01net)","publishedDate":"2026-06-02","accessedDate":"2026-10-22","tier":"T2","locator":"Release: Credo sells copper and optical interconnect products, retimers and DSPs only"}},"networking-silicon":{"low":2.584,"central":2.584,"high":2.584,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2026 (to 2 May 2026)","periodEnd":"2026-05-02","source":{"url":"https://www.01net.it/credo-technology-group-holding-ltd-reports-fourth-quarter-and-fiscal-year-2026-financial-results/","title":"Credo Technology Group Holding Ltd reports fourth quarter and fiscal year 2026 financial results","publisher":"Credo (press release, via 01net)","publishedDate":"2026-06-02","accessedDate":"2026-10-22","tier":"T5","locator":"Fiscal 2026 revenue $1,335,116 thousand"},"note":"T5 derivation (full inputs). Credo fiscal 2026 revenue $1,335.1M (active electrical cables, optical DSPs, retimers and SerDes chiplets); about 90% ($1.2B) taken as AI networking. 2025 value about $1.2B. Function total $46.44B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $35.1–47.1B (central $44.8B)). Share = 1.2 / 46.44 = 2.584%. Point value: the engine applies the T5 width. Itemisation and 2024 prior in a local evidence capture."},"host-cpus":{"participates":false,"evidence":{"url":"https://www.01net.it/credo-technology-group-holding-ltd-reports-fourth-quarter-and-fiscal-year-2026-financial-results/","title":"Credo Technology Group Holding Ltd reports fourth quarter and fiscal year 2026 financial results","publisher":"Credo (press release, via 01net)","publishedDate":"2026-06-02","accessedDate":"2026-10-22","tier":"T2","locator":"Release: Credo sells copper and optical interconnect products, retimers and DSPs only"}},"processor-ip":{"participates":false,"evidence":{"url":"https://www.01net.it/credo-technology-group-holding-ltd-reports-fourth-quarter-and-fiscal-year-2026-financial-results/","title":"Credo Technology Group Holding Ltd reports fourth quarter and fiscal year 2026 financial results","publisher":"Credo (press release, via 01net)","publishedDate":"2026-06-02","accessedDate":"2026-10-22","tier":"T2","locator":"Release: Credo sells copper and optical interconnect products, retimers and DSPs only"}}},"priorShares":{"networking-silicon":{"low":1.844,"central":1.844,"high":1.844,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"FY2025 (to 3 May 2025)","periodEnd":"2025-05-03","source":{"url":"https://www.01net.it/credo-technology-group-holding-ltd-reports-fourth-quarter-and-fiscal-year-2026-financial-results/","title":"Credo Technology Group Holding Ltd reports fourth quarter and fiscal year 2026 financial results","publisher":"Credo (press release, via 01net)","publishedDate":"2026-06-02","accessedDate":"2026-10-22","tier":"T5","locator":"Fiscal 2025 revenue $436,775 thousand (comparison column)"},"note":"T5 derivation of the prior-year share. Credo fiscal 2025 revenue $436.8M; about 90% ($0.39B) taken as AI networking, the same ratio as fiscal 2026. 2024 value about $0.3931B. Function total $21.32B = the itemised 2024 values of every candidate plus residual others, built like the 2025 total; candidates without a reported 2024 value are set at half their 2025 value (an assumption: most grew at least 100%). Share = 0.3931 / 21.32 = 1.844%. Point value: the engine applies the T5 width. Itemisation in a local evidence capture."}},"activityGrowth":{"current":{"low":1335116000,"central":1335116000,"high":1335116000,"unit":"USD","basis":"Total revenue (AECs, optical DSPs, retimers, chiplets), fiscal year, US dollars","period":"FY2026 (to 2 May 2026)","periodEnd":"2026-05-02","source":{"url":"https://www.01net.it/credo-technology-group-holding-ltd-reports-fourth-quarter-and-fiscal-year-2026-financial-results/","title":"Credo Technology Group Holding Ltd reports fourth quarter and fiscal year 2026 financial results","publisher":"Credo (press release, via 01net)","publishedDate":"2026-06-02","accessedDate":"2026-10-22","tier":"T2","locator":"Fiscal 2026 revenue $1,335,116 thousand"},"note":"Total revenue; about 90% is AI networking."},"prior":{"low":436775000,"central":436775000,"high":436775000,"unit":"USD","basis":"Total revenue (AECs, optical DSPs, retimers, chiplets), fiscal year, US dollars","period":"FY2025 (to 3 May 2025)","periodEnd":"2025-05-03","source":{"url":"https://www.01net.it/credo-technology-group-holding-ltd-reports-fourth-quarter-and-fiscal-year-2026-financial-results/","title":"Credo Technology Group Holding Ltd reports fourth quarter and fiscal year 2026 financial results","publisher":"Credo (press release, via 01net)","publishedDate":"2026-06-02","accessedDate":"2026-10-22","tier":"T2","locator":"Fiscal 2025 revenue $436,775 thousand (comparison column)"}}},"baselineRevenueUsd":{"low":1335116000,"central":1335116000,"high":1335116000,"unit":"USD","basis":"Total revenue, fiscal year, US dollars","period":"FY2026 (to 2 May 2026)","periodEnd":"2026-05-02","source":{"url":"https://www.01net.it/credo-technology-group-holding-ltd-reports-fourth-quarter-and-fiscal-year-2026-financial-results/","title":"Credo Technology Group Holding Ltd reports fourth quarter and fiscal year 2026 financial results","publisher":"Credo (press release, via 01net)","publishedDate":"2026-06-02","accessedDate":"2026-10-22","tier":"T2","locator":"Fiscal 2026 revenue $1,335,116 thousand"},"note":"Total revenue; about 90% is AI networking."},"screening":{"networking-silicon":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"ae89020540c2482fa2aac5cad05cbb5ccdf185a5974187fda4786c146e54ea97"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"ae89020540c2482fa2aac5cad05cbb5ccdf185a5974187fda4786c146e54ea97"}}},"meta":{"schema":"5lc-index-inputs/1","key":"chips-chip-design","entityId":"meta","name":"Meta","kind":"public","commercialActivityVerified":true,"shares":{"merchant-accelerators":{"participates":false,"evidence":{"url":"https://futurumgroup.com/insights/metas-mtia-partnership-with-broadcom-solidifies-the-future-of-xpus-in-inference-optimization/","title":"Meta's MTIA partnership with Broadcom solidifies the future of XPUs in inference optimization","publisher":"Futurum (Brendan Burke)","publishedDate":"2026-04-20","accessedDate":"2026-10-22","tier":"T3","locator":"Article: MTIA chips are for Meta's own data centres; Meta does not sell accelerators"}},"custom-accelerators":{"low":3.244,"central":4.866,"high":6.488,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://futurumgroup.com/insights/metas-mtia-partnership-with-broadcom-solidifies-the-future-of-xpus-in-inference-optimization/","title":"Meta's MTIA partnership with Broadcom solidifies the future of XPUs in inference optimization","publisher":"Futurum (Brendan Burke)","publishedDate":"2026-04-20","accessedDate":"2026-10-22","tier":"T5","locator":"Broadcom co-develops MTIA generations; MTIA runs ranking and recommendation"},"note":"T5 derivation (full inputs). MTIA is designed with Broadcom: Meta is taken as about 15% (range 10–20%) of Broadcom's roughly $12.5B of fiscal 2025 XPU revenue, a $1.875B programme with Meta's owner half $0.94B. No source gives MTIA revenue. 2025 value about $0.9375B. Function total $19.27B (was $17.03B before Amazon's programme was re-derived from chip volume, a recorded method decision; see fixes/a recorded method decision/itemisation-changes-a recorded method decision.md) = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $14.7–20.5B (central $16.5B)). Share = 0.9375 / 19.27 = 4.866%. Value range $0.625-1.25B gives the share range 3.244-6.488%. Volume cross-check (a recorded method decision): Tom's Hardware (21 May 2026, https://www.tomshardware.com/tech-industry/semiconductors/custom-ai-asics-examined-from-broadcom-to-mtia) says Meta 'has deployed hundreds of thousands of MTIA chips' (cumulative, no year) and TrendForce (4 August 2025, https://www.trendforce.com/news/2025/08/04/news-meta-reportedly-teams-up-with-broadcom-taps-quanta-for-next-gen-asic-powered-ai-servers/) says MTIA shipments are expected to double by 2026, without a number. At the TPU-implied $3,066 per chip, the $1.875B programme is about 0.6 million chips, inside 'hundreds of thousands'; no opened source gives 2025 volume, so the value is unchanged. Itemisation and 2024 prior in a local evidence capture."},"networking-silicon":{"low":0,"central":0.0646,"high":0.2153,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://futurumgroup.com/insights/metas-mtia-partnership-with-broadcom-solidifies-the-future-of-xpus-in-inference-optimization/","title":"Meta's MTIA partnership with Broadcom solidifies the future of XPUs in inference optimization","publisher":"Futurum (Brendan Burke)","publishedDate":"2026-04-20","accessedDate":"2026-10-22","tier":"T5","locator":"Broadcom provides Ethernet networking for Meta's clusters"},"note":"T5 derivation (full inputs). Meta-designed NICs used in its AI clusters, estimated at $0–0.1B. 2025 value about $0.03B. Function total $46.44B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $35.1–47.1B (central $44.8B)). Share = 0.03 / 46.44 = 0.0646%. Value range $0-0.1B gives the share range 0-0.2153%. Itemisation and 2024 prior in a local evidence capture."},"host-cpus":{"participates":false,"evidence":{"url":"https://futurumgroup.com/insights/metas-mtia-partnership-with-broadcom-solidifies-the-future-of-xpus-in-inference-optimization/","title":"Meta's MTIA partnership with Broadcom solidifies the future of XPUs in inference optimization","publisher":"Futurum (Brendan Burke)","publishedDate":"2026-04-20","accessedDate":"2026-10-22","tier":"T3","locator":"Article: Meta's chip programme covers MTIA accelerators; no Meta-designed host processor was in service in 2025"}},"processor-ip":{"participates":false,"evidence":{"url":"https://futurumgroup.com/insights/metas-mtia-partnership-with-broadcom-solidifies-the-future-of-xpus-in-inference-optimization/","title":"Meta's MTIA partnership with Broadcom solidifies the future of XPUs in inference optimization","publisher":"Futurum (Brendan Burke)","publishedDate":"2026-04-20","accessedDate":"2026-10-22","tier":"T3","locator":"Article: Meta does not license processor cores"}}},"screening":{"custom-accelerators":{"levelLow":1,"levelHigh":3},"networking-silicon":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"873f28f6d9cb9408f753c56f0ab79a5015d5aa69e51b0ce52b9e4edd24cf5d4e"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"873f28f6d9cb9408f753c56f0ab79a5015d5aa69e51b0ce52b9e4edd24cf5d4e"}}},"bytedance":{"schema":"5lc-index-inputs/1","key":"chips-chip-design","entityId":"bytedance","name":"ByteDance","kind":"private","commercialActivityVerified":false,"shares":{"merchant-accelerators":{"participates":false,"evidence":{"url":"https://www.trendforce.com/news/2024/06/24/news-broadcom-reportedly-developing-5nm-ai-chip-with-bytedance-manufactured-by-tsmc/","title":"[News] Broadcom Reportedly Developing 5nm AI Chip with ByteDance, Manufactured by TSMC","publisher":"TrendForce News (reporting Reuters and The Information)","publishedDate":"2024-06-24","accessedDate":"2026-10-22","tier":"T3","locator":"Reuters report as summarised by TrendForce: the Broadcom chip is meant to give ByteDance a steady supply of high-end chips for its own algorithms and apps (TikTok, Doubao); ByteDance buys NVIDIA chips; no report of ByteDance selling accelerators"}},"custom-accelerators":{"low":3.244,"central":3.244,"high":3.244,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.trendforce.com/news/2024/06/24/news-broadcom-reportedly-developing-5nm-ai-chip-with-bytedance-manufactured-by-tsmc/","title":"[News] Broadcom Reportedly Developing 5nm AI Chip with ByteDance, Manufactured by TSMC","publisher":"TrendForce News (reporting Reuters and The Information)","publishedDate":"2024-06-24","accessedDate":"2026-10-22","tier":"T5","locator":"Reuters, citing sources (24 June 2024): ByteDance and its existing partner Broadcom are developing a 5nm AI ASIC to be made by TSMC; design started, no tape-out yet"},"note":"T5 judgment (fix pass), not a sourced figure. No opened source gives ByteDance's 2025 volume with Broadcom. What the sources show: ByteDance is Broadcom's customer for custom AI ASICs ('existing partner', Reuters of 24 June 2024 via TrendForce News, https://www.trendforce.com/news/2024/06/24/news-broadcom-reportedly-developing-5nm-ai-chip-with-bytedance-manufactured-by-tsmc/); a 5nm chip was still before tape-out in mid-2024 and TSMC would not start making it in 2024, so the 2025 volume is uncertain; Reuters (11 February 2026, via https://the-decoder.com/bytedance-turns-to-samsung-for-custom-ai-chip-production-and-scarce-memory-supplies/) reports a separate in-house inference chip (SeedChip) planned at 100,000+ units in 2026, which ByteDance calls inaccurate. Judgment: ByteDance about 10% of Broadcom's XPU revenue, roughly $12.5B in fiscal 2025 (Broadcom AI revenue of $20B at 60-65% XPU, see Broadcom's file), a $1.25B programme with ByteDance's owner half $0.625B (the spread of 5-15% is the judgment's uncertainty; the engine applies the T5 width to the point value). An unsourced web compilation (Lambda Finance, 13 June 2026; not used as a source) also puts ByteDance at about 10% of Broadcom XPU revenue. 2025 value about $0.625B. Function total $19.27B (was $17.03B before Amazon's programme was re-derived from chip volume, a recorded method decision; see fixes/a recorded method decision/itemisation-changes-a recorded method decision.md) = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $14.7–20.5B (central $16.5B)). Share = 0.625 / 19.27 = 3.244%. Itemisation and 2024 prior in a local evidence capture (adjustments made in the fix pass: full-inputs/chips-chip-design/fixes/itemisation-changes.md)."},"networking-silicon":{"participates":false,"evidence":{"url":"https://the-decoder.com/bytedance-turns-to-samsung-for-custom-ai-chip-production-and-scarce-memory-supplies/","title":"ByteDance turns to Samsung for custom AI chip production and scarce memory supplies","publisher":"The Decoder (reporting Reuters)","publishedDate":"2026-02-11","accessedDate":"2026-10-22","tier":"T3","locator":"Reuters report as relayed by The Decoder: ByteDance's chip work is an in-house inference AI chip (SeedChip) and AI procurement; no networking-chip product is mentioned in this or the TrendForce report"}},"host-cpus":{"participates":false,"evidence":{"url":"https://the-decoder.com/bytedance-turns-to-samsung-for-custom-ai-chip-production-and-scarce-memory-supplies/","title":"ByteDance turns to Samsung for custom AI chip production and scarce memory supplies","publisher":"The Decoder (reporting Reuters)","publishedDate":"2026-02-11","accessedDate":"2026-10-22","tier":"T3","locator":"Reuters report as relayed by The Decoder: ByteDance's chip work is an in-house inference AI chip (SeedChip) and AI procurement; no host-processor product is mentioned in this or the TrendForce report"}},"processor-ip":{"participates":false,"evidence":{"url":"https://the-decoder.com/bytedance-turns-to-samsung-for-custom-ai-chip-production-and-scarce-memory-supplies/","title":"ByteDance turns to Samsung for custom AI chip production and scarce memory supplies","publisher":"The Decoder (reporting Reuters)","publishedDate":"2026-02-11","accessedDate":"2026-10-22","tier":"T3","locator":"Reuters report as relayed by The Decoder: ByteDance's chip work is an in-house inference AI chip (SeedChip) and AI procurement; no processor-core licensing is mentioned in this or the TrendForce report"}}},"screening":{"custom-accelerators":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"3712c4cfbcdf9da27d258ede62c3b6bb86c4f7e76b665261575e379317c6e2af"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"3712c4cfbcdf9da27d258ede62c3b6bb86c4f7e76b665261575e379317c6e2af"}}},"astera-labs":{"schema":"5lc-index-inputs/1","key":"chips-chip-design","entityId":"astera-labs","name":"Astera Labs","kind":"public","commercialActivityVerified":true,"shares":{"merchant-accelerators":{"participates":false,"evidence":{"url":"https://futurumgroup.com/insights/astera-labs-q4-2025-earnings-diversified-ai-connectivity-momentum","title":"Astera Labs Q4 2025 earnings: diversified AI connectivity momentum","publisher":"Futurum (Brendan Burke)","publishedDate":"2026-02-13","accessedDate":"2026-10-22","tier":"T2","locator":"Article: Astera sells connectivity products (switches, retimers, cable modules, CXL) only"}},"custom-accelerators":{"participates":false,"evidence":{"url":"https://futurumgroup.com/insights/astera-labs-q4-2025-earnings-diversified-ai-connectivity-momentum","title":"Astera Labs Q4 2025 earnings: diversified AI connectivity momentum","publisher":"Futurum (Brendan Burke)","publishedDate":"2026-02-13","accessedDate":"2026-10-22","tier":"T2","locator":"Article: Astera sells connectivity products (switches, retimers, cable modules, CXL) only"}},"networking-silicon":{"low":1.723,"central":1.723,"high":1.723,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.storagenewsletter.com/2026/02/25/astera-labs-fiscal-4q25-and-fy25-financial-results/","title":"Astera Labs: fiscal 4Q25 and FY25 financial results (reprint of Astera's press release)","publisher":"StorageNewsletter (reproducing Astera Labs' release)","publishedDate":"2026-02-25","accessedDate":"2026-10-22","tier":"T5","locator":"Record full-year revenue of $852.5 million, up 115% YoY"},"note":"T5 derivation (full inputs). Astera Labs' 2025 revenue was $852.5M, up 115% (Q4 $270.6M; company release, reprinted by StorageNewsletter); retimers, smart fabric switches, cable modules and CXL; about 94% ($0.8B) taken as AI networking. 2025 value about $0.8B. Function total $46.44B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $35.1–47.1B (central $44.8B)). Share = 0.8 / 46.44 = 1.723%. Point value: the engine applies the T5 width. Itemisation and 2024 prior in a local evidence capture."},"host-cpus":{"participates":false,"evidence":{"url":"https://futurumgroup.com/insights/astera-labs-q4-2025-earnings-diversified-ai-connectivity-momentum","title":"Astera Labs Q4 2025 earnings: diversified AI connectivity momentum","publisher":"Futurum (Brendan Burke)","publishedDate":"2026-02-13","accessedDate":"2026-10-22","tier":"T2","locator":"Article: Astera sells connectivity products (switches, retimers, cable modules, CXL) only"}},"processor-ip":{"participates":false,"evidence":{"url":"https://futurumgroup.com/insights/astera-labs-q4-2025-earnings-diversified-ai-connectivity-momentum","title":"Astera Labs Q4 2025 earnings: diversified AI connectivity momentum","publisher":"Futurum (Brendan Burke)","publishedDate":"2026-02-13","accessedDate":"2026-10-22","tier":"T2","locator":"Article: Astera sells connectivity products (switches, retimers, cable modules, CXL) only"}}},"priorShares":{"networking-silicon":{"low":1.748,"central":1.748,"high":1.748,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2024","periodEnd":"2024-12-31","source":{"url":"https://www.storagenewsletter.com/2026/02/25/astera-labs-fiscal-4q25-and-fy25-financial-results/","title":"Astera Labs: fiscal 4Q25 and FY25 financial results (reprint of Astera's press release)","publisher":"StorageNewsletter (reproducing Astera Labs' release)","publishedDate":"2026-02-25","accessedDate":"2026-10-22","tier":"T5","locator":"Full-year 2025 revenue $852.5 million, up 115%"},"note":"T5 derivation of the prior-year share. Astera Labs 2024 revenue about $396.5M ($852.5M / 2.15, from the reported 115% growth); about 94% ($0.37B) taken as AI networking, the same ratio as 2025. 2024 value about $0.3727B. Function total $21.32B = the itemised 2024 values of every candidate plus residual others, built like the 2025 total; candidates without a reported 2024 value are set at half their 2025 value (an assumption: most grew at least 100%). Share = 0.3727 / 21.32 = 1.748%. Point value: the engine applies the T5 width. Itemisation in a local evidence capture."}},"activityGrowth":{"current":{"low":852500000,"central":852500000,"high":852500000,"unit":"USD","basis":"Total revenue (retimers, switches, cable modules, CXL), calendar year, US dollars","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.storagenewsletter.com/2026/02/25/astera-labs-fiscal-4q25-and-fy25-financial-results/","title":"Astera Labs: fiscal 4Q25 and FY25 financial results (reprint of Astera's press release)","publisher":"StorageNewsletter (reproducing Astera Labs' release)","publishedDate":"2026-02-25","accessedDate":"2026-10-22","tier":"T2","locator":"Record full-year revenue of $852.5 million, up 115% YoY"},"note":"The 2024 value is derived from the reported 115% growth (rounded to the nearest $0.1M)."},"prior":{"low":396500000,"central":396500000,"high":396500000,"unit":"USD","basis":"Total revenue (retimers, switches, cable modules, CXL), calendar year, US dollars","period":"CY2024","periodEnd":"2024-12-31","source":{"url":"https://www.storagenewsletter.com/2026/02/25/astera-labs-fiscal-4q25-and-fy25-financial-results/","title":"Astera Labs: fiscal 4Q25 and FY25 financial results (reprint of Astera's press release)","publisher":"StorageNewsletter (reproducing Astera Labs' release)","publishedDate":"2026-02-25","accessedDate":"2026-10-22","tier":"T2","locator":"Derived: $852.5M / (1 + 115%)"},"note":"Derived, not reported: 2025 revenue of $852.5M divided by (1 + 115%), the growth rate in Astera Labs' release; rounded to the nearest $0.1M."}},"baselineRevenueUsd":{"low":852500000,"central":852500000,"high":852500000,"unit":"USD","basis":"Total revenue, calendar year, US dollars","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.storagenewsletter.com/2026/02/25/astera-labs-fiscal-4q25-and-fy25-financial-results/","title":"Astera Labs: fiscal 4Q25 and FY25 financial results (reprint of Astera's press release)","publisher":"StorageNewsletter (reproducing Astera Labs' release)","publishedDate":"2026-02-25","accessedDate":"2026-10-22","tier":"T2","locator":"Record full-year revenue of $852.5 million"},"note":"Total revenue."},"screening":{"networking-silicon":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"ceceb19880db48b090a3a84b2258ba46f7d50a320c498438c3c35efe58e1e432"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"ceceb19880db48b090a3a84b2258ba46f7d50a320c498438c3c35efe58e1e432"}}},"hygon":{"schema":"5lc-index-inputs/1","key":"chips-chip-design","entityId":"hygon","name":"Hygon","kind":"public","commercialActivityVerified":true,"shares":{"merchant-accelerators":{"low":0.1707,"central":0.2845,"high":0.4552,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://kfgo.com/?p=1285523","title":"Chinese chipmakers claim nearly half of local market as Nvidia's lead shrinks (IDC data)","publisher":"Reuters (Che Pan and Laurie Chen), via KFGO","publishedDate":"2026-04-01","accessedDate":"2026-10-22","tier":"T5","locator":"IDC 2025: Hygon 5% of Chinese-vendor shipments"},"note":"T5 derivation (full inputs). IDC puts Hygon at 5% of the 1.65 million Chinese-vendor cards (about 82,500) in 2025 (Reuters); at $3,600–9,700 each the DCU value is $0.3–0.8B (central $0.5B). Hygon's 2025 revenue was CNY 14.377B (https://www.cls.cn/detail/2337024) with no CPU/DCU split. 2025 value about $0.5B. Function total $175.74B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $168.7–183.3B (central $175.8B)). Share = 0.5 / 175.74 = 0.2845%. Value range $0.3-0.8B gives the share range 0.1707-0.4552%. Itemisation and 2024 prior in a local evidence capture."},"custom-accelerators":{"participates":false,"evidence":{"url":"https://www.cls.cn/detail/2337024","title":"海光信息2025年年报：营业收入143.77亿元 (Hygon 2025 annual report)","publisher":"财联社 (Cailian Press)","publishedDate":"2026-04-07","accessedDate":"2026-10-22","tier":"T2","locator":"Annual report coverage: Hygon sells its own CPU and DCU products; no custom design for other owners"}},"networking-silicon":{"participates":false,"evidence":{"url":"https://www.cls.cn/detail/2337024","title":"海光信息2025年年报：营业收入143.77亿元 (Hygon 2025 annual report)","publisher":"财联社 (Cailian Press)","publishedDate":"2026-04-07","accessedDate":"2026-10-22","tier":"T2","locator":"Annual report coverage: Hygon's products are high-end processors (CPU and DCU); no networking chips"}},"host-cpus":{"low":1.496,"central":2.792,"high":4.487,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.cls.cn/detail/2337024","title":"海光信息2025年年报：营业收入143.77亿元 (Hygon 2025 annual report)","publisher":"财联社 (Cailian Press)","publishedDate":"2026-04-07","accessedDate":"2026-10-22","tier":"T5","locator":"2025 revenue 143.77亿元; CPU and DCU products"},"note":"T5 derivation (full inputs). Hygon 2025 revenue CNY 14.377B ($2.0B at 7.18) less the DCU estimate leaves about $1.4B of x86 processors; 20% ($0.28B, range $0.15–0.45B) taken as AI-server hosts. 2025 value about $0.28B. Function total $10.03B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $7.0–12.9B (central $9.6B)). Share = 0.28 / 10.03 = 2.792%. Value range $0.15-0.45B gives the share range 1.496-4.487%. Itemisation and 2024 prior in a local evidence capture."},"processor-ip":{"participates":false,"evidence":{"url":"https://www.cls.cn/detail/2337024","title":"海光信息2025年年报：营业收入143.77亿元 (Hygon 2025 annual report)","publisher":"财联社 (Cailian Press)","publishedDate":"2026-04-07","accessedDate":"2026-10-22","tier":"T2","locator":"Annual report coverage: Hygon opened its HSL bus protocol to partners but reports no processor-core licensing revenue"}}},"priorShares":{"merchant-accelerators":{"low":0.134,"central":0.2845,"high":0.4465,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2024","periodEnd":"2024-12-31","source":{"url":"https://www.cls.cn/detail/2337024","title":"海光信息2025年年报：营业收入143.77亿元 (Hygon 2025 annual report)","publisher":"财联社 (Cailian Press)","publishedDate":"2026-04-07","accessedDate":"2026-10-22","tier":"T5","locator":"2025 revenue 143.77亿元, up 56.92%"},"note":"T5 derivation of the prior-year share. Hygon's revenue was CNY 14.377B in 2025, up 56.92%, so CNY 9.162B in 2024 ($1.27B at 7.19). The DCU part is scaled with revenue from the 2025 estimate ($0.5B x 9.162 / 14.377 = $0.32B); Hygon reports no DCU split. 2024 value about $0.3186B. Function total $111.97B = the itemised 2024 values of every candidate plus residual others, built like the 2025 total; candidates without a reported 2024 value are set at half their 2025 value (an assumption: most grew at least 100%). Share = 0.3186 / 111.97 = 0.2845%. Value range $0.15-0.5B gives the share range 0.134-0.4465%. Itemisation in a local evidence capture."},"host-cpus":{"low":1.816,"central":3.239,"high":5.447,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2024","periodEnd":"2024-12-31","source":{"url":"https://www.cls.cn/detail/2337024","title":"海光信息2025年年报：营业收入143.77亿元 (Hygon 2025 annual report)","publisher":"财联社 (Cailian Press)","publishedDate":"2026-04-07","accessedDate":"2026-10-22","tier":"T5","locator":"2025 revenue 143.77亿元, up 56.92%"},"note":"T5 derivation of the prior-year share. x86 processors scaled with revenue from the 2025 estimate ($0.28B x 9.162 / 14.377 = $0.18B, range $0.10-0.30B). 2024 value about $0.1784B. Function total $5.51B = the itemised 2024 values of every candidate plus residual others, built like the 2025 total; candidates without a reported 2024 value are set at half their 2025 value (an assumption: most grew at least 100%). Share = 0.1784 / 5.51 = 3.239%. Value range $0.1-0.3B gives the share range 1.816-5.447%. Itemisation in a local evidence capture."}},"activityGrowth":{"current":{"low":14377000000,"central":14377000000,"high":14377000000,"unit":"CNY","basis":"Total revenue (x86 processors and DCU accelerators), calendar year, Chinese yuan","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.cls.cn/detail/2337024","title":"海光信息2025年年报：营业收入143.77亿元 (Hygon 2025 annual report)","publisher":"财联社 (Cailian Press)","publishedDate":"2026-04-07","accessedDate":"2026-10-22","tier":"T2","locator":"2025 revenue 143.77亿元"},"note":"Hygon reports no DCU split, so total revenue (all data-centre processors and accelerators) is used."},"prior":{"low":9161993372,"central":9161993372,"high":9161993372,"unit":"CNY","basis":"Total revenue (x86 processors and DCU accelerators), calendar year, Chinese yuan","period":"CY2024","periodEnd":"2024-12-31","source":{"url":"https://www.cls.cn/detail/2337024","title":"海光信息2025年年报：营业收入143.77亿元 (Hygon 2025 annual report)","publisher":"财联社 (Cailian Press)","publishedDate":"2026-04-07","accessedDate":"2026-10-22","tier":"T2","locator":"Derived: 14.377B / (1 + 56.92%)"},"note":"Derived, not reported: 2025 revenue of CNY 14.377B divided by (1 + 56.92%), the growth rate reported by Cailian Press (annual report)."}},"committedExpansion":{"low":0.1404,"central":0.1404,"high":0.1404,"unit":"ratio","basis":"Contract liabilities ÷ 2025 total revenue","period":"31 December 2025","periodEnd":"2025-12-31","source":{"url":"https://www.cls.cn/detail/2337024","title":"海光信息2025年年报：营业收入143.77亿元 (Hygon 2025 annual report)","publisher":"财联社 (Cailian Press)","publishedDate":"2026-04-07","accessedDate":"2026-10-22","tier":"T2","locator":"Annual results: contract liabilities (合同负债) of 20.18亿元 at year-end 2025, up 123.42%"},"note":"Contract liabilities (合同负债) were CNY 2.018B at the end of 2025, up 123.42% in the year (company says customers prepaid more on contracts); revenue was CNY 14.377B; 2.018 / 14.377 = 0.1404. Both numbers cover all of Hygon's products (the company reports no DCU split)."},"baselineRevenueUsd":{"low":2002400000,"central":2002400000,"high":2002400000,"unit":"USD","basis":"Total revenue (x86 processors and DCU accelerators), calendar year, US dollars","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.cls.cn/detail/2337024","title":"海光信息2025年年报：营业收入143.77亿元 (Hygon 2025 annual report)","publisher":"财联社 (Cailian Press)","publishedDate":"2026-04-07","accessedDate":"2026-10-22","tier":"T2","locator":"2025 revenue 143.77亿元"},"note":"CNY 14.377B at about 7.18 CNY per USD; Hygon reports no DCU split, so this is all of its data-centre processor and accelerator revenue."},"screening":{"merchant-accelerators":{"levelLow":1,"levelHigh":3},"host-cpus":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"0b25d172c5d7dc79e028c248e26b2697096080a1c933ee5feaa9a7298506b5d5"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"0b25d172c5d7dc79e028c248e26b2697096080a1c933ee5feaa9a7298506b5d5"}}},"baidu":{"schema":"5lc-index-inputs/1","key":"chips-chip-design","entityId":"baidu","name":"Baidu","kind":"public","commercialActivityVerified":true,"shares":{"merchant-accelerators":{"low":0.1387,"central":0.1526,"high":0.1941,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.trendforce.com/news/2026/05/08/news-baidu-chip-unit-kunlunxin-reportedly-launches-star-market-ipo-process-hk-listing-valuation-seen-near-hk100b/","title":"Baidu chip unit Kunlunxin reportedly launches STAR Market IPO process","publisher":"TrendForce News","publishedDate":"2026-05-08","accessedDate":"2026-10-22","tier":"T5","locator":"Kunlunxin 2025 revenue expected above RMB 3.5 billion; external customers over 50%"},"note":"T5 derivation (full inputs). Kunlunxin chips sold to outside customers. Kunlunxin (a Baidu subsidiary) expected 2025 revenue above CNY 3.5B ($0.4875B at 7.18), with outside customers more than half (investment materials reported by Reuters, via TrendForce); IDC counts about 116,000 Kunlunxin cards (https://kfgo.com/?p=1285523). Outside sales are taken as 55% (range 50-70%) of the revenue. 2025 value about $0.2681B. Function total $175.74B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $168.7–183.3B (central $175.8B)). Share = 0.2681 / 175.74 = 0.1526%. Value range $0.244-0.341B gives the share range 0.1387-0.1941%. Itemisation and 2024 prior in a local evidence capture."},"custom-accelerators":{"low":0.7589,"central":1.139,"high":1.265,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.trendforce.com/news/2026/05/08/news-baidu-chip-unit-kunlunxin-reportedly-launches-star-market-ipo-process-hk-listing-valuation-seen-near-hk100b/","title":"Baidu chip unit Kunlunxin reportedly launches STAR Market IPO process","publisher":"TrendForce News","publishedDate":"2026-05-08","accessedDate":"2026-10-22","tier":"T5","locator":"Kunlunxin sells to Baidu and outside customers"},"note":"T5 derivation (full inputs). Kunlunxin chips that Baidu uses itself: 45% (range 30-50%) of Kunlunxin's expected 2025 revenue of more than CNY 3.5B ($0.4875B at 7.18), valued at the owner's cost; Kunlunxin is Baidu's controlled subsidiary, so the owner and the design partner are the same group and the whole programme value is credited to Baidu. This replaces the screening's zero, which counted these chips under merchant accelerators. 2025 value about $0.2194B. Function total $19.27B (was $17.03B before Amazon's programme was re-derived from chip volume, a recorded method decision; see fixes/a recorded method decision/itemisation-changes-a recorded method decision.md) = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $14.7–20.5B (central $16.5B)). Share = 0.2194 / 19.27 = 1.139%. Value range $0.146-0.244B gives the share range 0.7589-1.265%. Itemisation and 2024 prior in a local evidence capture."},"networking-silicon":{"low":0,"central":0.02153,"high":0.08614,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.trendforce.com/news/2026/05/08/news-baidu-chip-unit-kunlunxin-reportedly-launches-star-market-ipo-process-hk-listing-valuation-seen-near-hk100b/","title":"Baidu chip unit Kunlunxin reportedly launches STAR Market IPO process","publisher":"TrendForce News","publishedDate":"2026-05-08","accessedDate":"2026-10-22","tier":"T5","locator":"Kunlunxin products and Tianchi supernodes"},"note":"T5 derivation (full inputs). Baidu-designed network chips in AI clusters, estimated at $0–0.04B. 2025 value about $0.01B. Function total $46.44B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $35.1–47.1B (central $44.8B)). Share = 0.01 / 46.44 = 0.02153%. Value range $0-0.04B gives the share range 0-0.08614%. Itemisation and 2024 prior in a local evidence capture."},"host-cpus":{"participates":false,"evidence":{"url":"https://www.trendforce.com/news/2026/05/08/news-baidu-chip-unit-kunlunxin-reportedly-launches-star-market-ipo-process-hk-listing-valuation-seen-near-hk100b/","title":"Baidu chip unit Kunlunxin reportedly launches STAR Market IPO process","publisher":"TrendForce News","publishedDate":"2026-05-08","accessedDate":"2026-10-22","tier":"T3","locator":"Article: Kunlunxin's roadmap covers AI accelerators (M100, M300) and Tianchi supernodes; no host processor"}},"processor-ip":{"participates":false,"evidence":{"url":"https://www.trendforce.com/news/2026/05/08/news-baidu-chip-unit-kunlunxin-reportedly-launches-star-market-ipo-process-hk-listing-valuation-seen-near-hk100b/","title":"Baidu chip unit Kunlunxin reportedly launches STAR Market IPO process","publisher":"TrendForce News","publishedDate":"2026-05-08","accessedDate":"2026-10-22","tier":"T3","locator":"Article: no processor-core licensing"}}},"activityGrowth":{"current":{"low":3500000000,"central":3500000000,"high":3500000000,"unit":"CNY","basis":"Kunlunxin revenue (AI accelerators sold to Baidu and outside customers), calendar year, Chinese yuan; 2025 is an expected figure","period":"CY2025 (expected)","periodEnd":"2025-12-31","source":{"url":"https://www.trendforce.com/news/2026/05/08/news-baidu-chip-unit-kunlunxin-reportedly-launches-star-market-ipo-process-hk-listing-valuation-seen-near-hk100b/","title":"Baidu chip unit Kunlunxin reportedly launches STAR Market IPO process","publisher":"TrendForce News","publishedDate":"2026-05-08","accessedDate":"2026-10-22","tier":"T3","locator":"Expected to increase revenue to more than RMB 3.5 billion in 2025"},"note":"The 2025 value is a forecast reported before the year ended (investment materials seen by Reuters) and is a floor ('more than'); the 2024 value is 'approximately' RMB 2B."},"prior":{"low":2000000000,"central":2000000000,"high":2000000000,"unit":"CNY","basis":"Kunlunxin revenue (AI accelerators sold to Baidu and outside customers), calendar year, Chinese yuan; 2025 is an expected figure","period":"CY2024","periodEnd":"2024-12-31","source":{"url":"https://www.trendforce.com/news/2026/05/08/news-baidu-chip-unit-kunlunxin-reportedly-launches-star-market-ipo-process-hk-listing-valuation-seen-near-hk100b/","title":"Baidu chip unit Kunlunxin reportedly launches STAR Market IPO process","publisher":"TrendForce News","publishedDate":"2026-05-08","accessedDate":"2026-10-22","tier":"T3","locator":"Kunlunxin generated revenue of approximately RMB 2 billion in 2024"}}},"baselineRevenueUsd":{"low":487500000,"central":487500000,"high":487500000,"unit":"USD","basis":"Kunlunxin revenue (unit), calendar year, US dollars; expected figure","period":"CY2025 (expected)","periodEnd":"2025-12-31","source":{"url":"https://www.trendforce.com/news/2026/05/08/news-baidu-chip-unit-kunlunxin-reportedly-launches-star-market-ipo-process-hk-listing-valuation-seen-near-hk100b/","title":"Baidu chip unit Kunlunxin reportedly launches STAR Market IPO process","publisher":"TrendForce News","publishedDate":"2026-05-08","accessedDate":"2026-10-22","tier":"T3","locator":"Expected to increase revenue to more than RMB 3.5 billion in 2025"},"note":"Kunlunxin only (a Baidu subsidiary), never Baidu's total; a forecast floor, CNY 3.5B at about 7.18 CNY per USD."},"screening":{"merchant-accelerators":{"levelLow":1,"levelHigh":3},"networking-silicon":{"levelLow":1,"levelHigh":2},"custom-accelerators":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"e5827bc8975a7f3539a7fcde6dfca3734f6653dacfe4f03219d1bd0257542083"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"e5827bc8975a7f3539a7fcde6dfca3734f6653dacfe4f03219d1bd0257542083"}}},"cambricon":{"schema":"5lc-index-inputs/1","key":"chips-chip-design","entityId":"cambricon","name":"Cambricon","kind":"public","commercialActivityVerified":true,"shares":{"merchant-accelerators":{"low":0.5128,"central":0.5128,"high":0.5128,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://static.cninfo.com.cn/finalpage/2026-03-13/1225007336.PDF","title":"中科寒武纪科技股份有限公司2025年年度报告 (Cambricon Technologies 2025 annual report)","publisher":"Cambricon Technologies (via cninfo)","publishedDate":"2026-03-12","accessedDate":"2026-10-22","tier":"T1","locator":"Note on revenue by product (营业收入分解), PDF p.178: cloud product line CNY 6,476,855,244.59; total revenue CNY 6,497,196,198.68"},"note":"T5 derivation (fix pass). Cambricon's cloud product line (AI accelerator cards and systems) earned CNY 6,476,855,244.59 in 2025 (annual report, revenue-by-product note), 99.7% of its CNY 6,497,196,198.68 revenue; National Business Daily reports the same figures (https://www.nbd.com.cn/articles/2026-03-12/4290381.html). Converted at 7.1875 CNY per USD (Federal Reserve G.5A (release of 5 January 2026, https://www.federalreserve.gov/releases/g5a/current/default.htm): 2025 average 7.1875 CNY per USD, 2024 average 7.1957): $0.9011B. IDC counts about 116,000 Cambricon cards in China in 2025 (Reuters, https://kfgo.com/?p=1285523). 2025 value about $0.9011B. Function total $175.74B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $168.7–183.3B (central $175.8B)). Share = 0.9011 / 175.74 = 0.5128%. Point value: the engine applies the T5 width. Itemisation and 2024 prior in a local evidence capture (adjustments made in the fix pass: full-inputs/chips-chip-design/fixes/itemisation-changes.md)."},"custom-accelerators":{"participates":false,"evidence":{"url":"https://static.cninfo.com.cn/finalpage/2026-03-13/1225007336.PDF","title":"中科寒武纪科技股份有限公司2025年年度报告 (Cambricon Technologies 2025 annual report)","publisher":"Cambricon Technologies (via cninfo)","publishedDate":"2026-03-12","accessedDate":"2026-10-22","tier":"T1","locator":"Section 3 (MD&A) I(1) main business: the company designs and sells AI core chips for cloud servers, edge devices and terminals; product lines are cloud, edge, and IP licensing and software (PDF p.12); revenue by product (note, PDF p.178) lists cloud, edge, IP licensing and software, and other (CNY 14.7M, 0.2%); no custom-chip or design-service line"}},"networking-silicon":{"participates":false,"evidence":{"url":"https://static.cninfo.com.cn/finalpage/2026-03-13/1225007336.PDF","title":"中科寒武纪科技股份有限公司2025年年度报告 (Cambricon Technologies 2025 annual report)","publisher":"Cambricon Technologies (via cninfo)","publishedDate":"2026-03-12","accessedDate":"2026-10-22","tier":"T1","locator":"Section 3 (MD&A) I(1) main business: the company designs and sells AI core chips for cloud servers, edge devices and terminals; product lines are cloud, edge, and IP licensing and software (PDF p.12); revenue by product (note, PDF p.178) lists cloud, edge, IP licensing and software, and other (CNY 14.7M, 0.2%); no networking-chip line"}},"host-cpus":{"participates":false,"evidence":{"url":"https://static.cninfo.com.cn/finalpage/2026-03-13/1225007336.PDF","title":"中科寒武纪科技股份有限公司2025年年度报告 (Cambricon Technologies 2025 annual report)","publisher":"Cambricon Technologies (via cninfo)","publishedDate":"2026-03-12","accessedDate":"2026-10-22","tier":"T1","locator":"Section 3 (MD&A) I(1) main business: the company designs and sells AI core chips for cloud servers, edge devices and terminals; product lines are cloud, edge, and IP licensing and software (PDF p.12); revenue by product (note, PDF p.178) lists cloud, edge, IP licensing and software, and other (CNY 14.7M, 0.2%); no general-purpose server-processor line"}},"processor-ip":{"low":0.0537,"central":0.0537,"high":0.0537,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://static.cninfo.com.cn/finalpage/2026-03-13/1225007336.PDF","title":"中科寒武纪科技股份有限公司2025年年度报告 (Cambricon Technologies 2025 annual report)","publisher":"Cambricon Technologies (via cninfo)","publishedDate":"2026-03-12","accessedDate":"2026-10-22","tier":"T1","locator":"Note on revenue by product (营业收入分解), PDF p.178: IP licensing and software CNY 2,288,749.93; main business (PDF p.12): IP licensing is licensing of the company's smart-processor IP, plus a base system-software platform"},"note":"T5 derivation (fix pass). Cambricon reports an 'IP licensing and software' line of CNY 2,288,749.93 in 2025 (about 0.04% of revenue; National Business Daily gives the same, CNY 2.29M), so the earlier 'does not participate' was wrong. The line combines licensing of the company's smart-processor IP with its system-software platform and the filing does not split it, and the licensed IP is not stated to be data-centre IP, so the whole line is taken as an upper bound for this function. Converted at 7.1875 CNY per USD (Federal Reserve G.5A (release of 5 January 2026, https://www.federalreserve.gov/releases/g5a/current/default.htm): 2025 average 7.1875 CNY per USD, 2024 average 7.1957): $0.318M. Function total $0.593B = the itemised 2025 values of every candidate plus residual others (the card's own total is $0.40–1.0B (central $0.59B)). Share = 0.000318 / 0.593 = 0.0537%. Below the 1% material threshold. Point value: the engine applies the T5 width. Itemisation and 2024 prior in a local evidence capture (adjustments made in the fix pass: full-inputs/chips-chip-design/fixes/itemisation-changes.md)."}},"priorShares":{"merchant-accelerators":{"low":0.1448,"central":0.1448,"high":0.1448,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2024","periodEnd":"2024-12-31","source":{"url":"https://static.cninfo.com.cn/finalpage/2025-04-19/1223155683.PDF","title":"中科寒武纪科技股份有限公司2024年年度报告 (Cambricon Technologies 2024 annual report)","publisher":"Cambricon Technologies (via cninfo)","publishedDate":"2025-04-18","accessedDate":"2026-10-22","tier":"T1","locator":"Note on revenue by product (营业收入分解), PDF p.195: cloud product line CNY 1,166,278,485.36; total revenue CNY 1,174,464,377.35"},"note":"T5 derivation of the prior-year share. Cambricon's cloud product line earned CNY 1,166,278,485.36 in 2024 (2024 annual report, revenue-by-product note; it equals the 2025 figure divided by 1 + 455.34%, the growth in the 2025 report). At 7.1957 CNY per USD (Federal Reserve G.5A (release of 5 January 2026, https://www.federalreserve.gov/releases/g5a/current/default.htm): 2025 average 7.1875 CNY per USD, 2024 average 7.1957) that is $0.1621B. 2024 value about $0.1621B. Function total $111.97B = the itemised 2024 values of every candidate plus residual others, built like the 2025 total; candidates without a reported 2024 value are set at half their 2025 value (an assumption: most grew at least 100%). Share = 0.1621 / 111.97 = 0.1448%. Point value: the engine applies the T5 width. Itemisation and 2024 prior in a local evidence capture (adjustments made in the fix pass: full-inputs/chips-chip-design/fixes/itemisation-changes.md)."},"processor-ip":{"low":0.0157,"central":0.0157,"high":0.0157,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2024","periodEnd":"2024-12-31","source":{"url":"https://static.cninfo.com.cn/finalpage/2025-04-19/1223155683.PDF","title":"中科寒武纪科技股份有限公司2024年年度报告 (Cambricon Technologies 2024 annual report)","publisher":"Cambricon Technologies (via cninfo)","publishedDate":"2025-04-18","accessedDate":"2026-10-22","tier":"T1","locator":"Note on revenue by product (营业收入分解), PDF p.195: IP licensing and software CNY 412,389.38"},"note":"T5 derivation of the prior-year share. The 2024 annual report shows an 'IP licensing and software' line of CNY 412,389.38; at 7.1957 CNY per USD (Federal Reserve G.5A (release of 5 January 2026, https://www.federalreserve.gov/releases/g5a/current/default.htm): 2025 average 7.1875 CNY per USD, 2024 average 7.1957) that is $0.057M, taken as an upper bound for processor IP, as in 2025. Function total $0.365B = the itemised 2024 values of every candidate plus residual others, built like the 2025 total. Share = 0.000057 / 0.365 = 0.0157%. Point value: the engine applies the T5 width. Itemisation and 2024 prior in a local evidence capture (adjustments made in the fix pass: full-inputs/chips-chip-design/fixes/itemisation-changes.md)."}},"activityGrowth":{"current":{"low":6476855244.59,"central":6476855244.59,"high":6476855244.59,"unit":"CNY","basis":"Cloud product line revenue (AI accelerator cards and systems), calendar year, Chinese yuan","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://static.cninfo.com.cn/finalpage/2026-03-13/1225007336.PDF","title":"中科寒武纪科技股份有限公司2025年年度报告 (Cambricon Technologies 2025 annual report)","publisher":"Cambricon Technologies (via cninfo)","publishedDate":"2026-03-12","accessedDate":"2026-10-22","tier":"T1","locator":"Note on revenue by product (营业收入分解), PDF p.178: cloud product line CNY 6,476,855,244.59"},"note":"Reported in the 2025 annual report; National Business Daily gives CNY 6.476B and growth of 455.34%."},"prior":{"low":1166278485.36,"central":1166278485.36,"high":1166278485.36,"unit":"CNY","basis":"Cloud product line revenue (AI accelerator cards and systems), calendar year, Chinese yuan","period":"CY2024","periodEnd":"2024-12-31","source":{"url":"https://static.cninfo.com.cn/finalpage/2025-04-19/1223155683.PDF","title":"中科寒武纪科技股份有限公司2024年年度报告 (Cambricon Technologies 2024 annual report)","publisher":"Cambricon Technologies (via cninfo)","publishedDate":"2025-04-18","accessedDate":"2026-10-22","tier":"T1","locator":"Note on revenue by product (营业收入分解), PDF p.195: cloud product line CNY 1,166,278,485.36"},"note":"Reported in the 2024 annual report (not derived): 6,476,855,244.59 / 1,166,278,485.36 - 1 = 455.34%, matching the growth the 2025 report states."}},"baselineRevenueUsd":{"low":901127686,"central":901127686,"high":901127686,"unit":"USD","basis":"Cloud product line revenue, calendar year, US dollars","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://static.cninfo.com.cn/finalpage/2026-03-13/1225007336.PDF","title":"中科寒武纪科技股份有限公司2025年年度报告 (Cambricon Technologies 2025 annual report)","publisher":"Cambricon Technologies (via cninfo)","publishedDate":"2026-03-12","accessedDate":"2026-10-22","tier":"T1","locator":"Note on revenue by product (营业收入分解), PDF p.178: cloud product line CNY 6,476,855,244.59"},"note":"CNY 6,476,855,244.59 converted at 7.1875 CNY per USD (Federal Reserve G.5A (release of 5 January 2026, https://www.federalreserve.gov/releases/g5a/current/default.htm): 2025 average 7.1875 CNY per USD, 2024 average 7.1957)."},"screening":{"merchant-accelerators":{"levelLow":1,"levelHigh":3},"processor-ip":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"6adfa650ee4366958430b285548ef13501108366c988fb700a94c32fc68c0639"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"6adfa650ee4366958430b285548ef13501108366c988fb700a94c32fc68c0639"}}},"cisco":{"schema":"5lc-index-inputs/1","key":"chips-chip-design","entityId":"cisco","name":"Cisco","kind":"public","commercialActivityVerified":true,"shares":{"merchant-accelerators":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/858877/000085887726000132/csco-20260725.htm","title":"Cisco Systems Form 10-K for the fiscal year ended July 25, 2026","publisher":"Cisco / U.S. SEC","publishedDate":"2026-09-02","accessedDate":"2026-10-22","tier":"T1","locator":"Item 1 Business: Cisco sells networking, security, collaboration and observability products; no AI accelerators"}},"custom-accelerators":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/858877/000085887726000132/csco-20260725.htm","title":"Cisco Systems Form 10-K for the fiscal year ended July 25, 2026","publisher":"Cisco / U.S. SEC","publishedDate":"2026-09-02","accessedDate":"2026-10-22","tier":"T1","locator":"Item 1 Business: no custom accelerator design"}},"networking-silicon":{"low":0.5017,"central":1.249,"high":2.5066,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025 (derived from fiscal 2025 and fiscal 2026)","periodEnd":"2025-12-31","source":{"url":"https://www.sec.gov/Archives/edgar/data/858877/000085887726000132/csco-20260725.htm","title":"Cisco Systems Form 10-K for the fiscal year ended July 25, 2026","publisher":"Cisco / U.S. SEC","publishedDate":"2026-09-02","accessedDate":"2026-10-22","tier":"T5","locator":"Item 7: AI infrastructure from hyperscalers ~6% of revenue in fiscal 2026 vs less than 2% in fiscal 2025; systems predominantly powered by Cisco Silicon One"},"note":"T5 derivation (fix pass: range restated from its inputs). Cisco sells Silicon One chips inside its own switches and routers, so this counts chip value only. Hyperscaler AI infrastructure was less than 2% of fiscal 2025 revenue (10-K), under about $1.13B at system level (2% x $56.7B; an upper bound), and about $4B in fiscal 2026 (earnings release, 12 August 2026: https://newsroom.cisco.com/c/r/newsroom/en/us/a/y2026/m08/cisco-reports-fourth-quarter-earnings.html). Calendar 2025 is taken as 7/12 of fiscal 2025 plus 5/12 of fiscal 2026: 7/12 x 1.134 + 5/12 x 4.0 = about $2.33B at system level. The chip is taken as 25% of system value at the centre (an assumption, not a Cisco figure) and as 10% to 50% at the ends of the range: $2.33B x 10% = $0.233B (low), x 25% = $0.58B (central, as itemised), x 50% = $1.164B (high). 2025 value about $0.58B. Function total $46.44B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $35.1–47.1B (central $44.8B)). Share = 0.58 / 46.44 = 1.249%. Value range $0.233-1.164B gives the share range 0.5017-2.5066%. Itemisation and 2024 prior in a local evidence capture (adjustments made in the fix pass: full-inputs/chips-chip-design/fixes/itemisation-changes.md)."},"host-cpus":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/858877/000085887726000132/csco-20260725.htm","title":"Cisco Systems Form 10-K for the fiscal year ended July 25, 2026","publisher":"Cisco / U.S. SEC","publishedDate":"2026-09-02","accessedDate":"2026-10-22","tier":"T1","locator":"Item 1 Business: Cisco's UCS servers use other companies' processors; no processors of its own"}},"processor-ip":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/858877/000085887726000132/csco-20260725.htm","title":"Cisco Systems Form 10-K for the fiscal year ended July 25, 2026","publisher":"Cisco / U.S. SEC","publishedDate":"2026-09-02","accessedDate":"2026-10-22","tier":"T1","locator":"Item 1 Business: no processor-core licensing"}}},"activityGrowth":{"current":{"low":4000000000,"central":4000000000,"high":4000000000,"unit":"USD","basis":"Hyperscaler AI infrastructure revenue (systems built on Cisco Silicon One and optics), fiscal year, system level, US dollars","period":"FY2026 (to 25 July 2026)","periodEnd":"2026-07-25","source":{"url":"https://newsroom.cisco.com/c/r/newsroom/en/us/a/y2026/m08/cisco-reports-fourth-quarter-earnings.html","title":"Cisco reports fourth quarter and fiscal year 2026 earnings","publisher":"Cisco","publishedDate":"2026-08-12","accessedDate":"2026-10-22","tier":"T2","locator":"Delivered approximately $4 billion of revenue in FY 2026"},"note":"The fiscal 2025 value is an upper bound (the 10-K says less than 2% of $56.7B revenue), so the growth shown is a lower bound; both are system-level revenue, not chip value."},"prior":{"low":1134000000,"central":1134000000,"high":1134000000,"unit":"USD","basis":"Hyperscaler AI infrastructure revenue (systems built on Cisco Silicon One and optics), fiscal year, system level, US dollars","period":"FY2025 (to 26 July 2025)","periodEnd":"2025-07-26","source":{"url":"https://www.sec.gov/Archives/edgar/data/858877/000085887726000132/csco-20260725.htm","title":"Cisco Systems Form 10-K for the fiscal year ended July 25, 2026","publisher":"Cisco / U.S. SEC","publishedDate":"2026-09-02","accessedDate":"2026-10-22","tier":"T1","locator":"Item 7: hyperscaler AI infrastructure was less than 2% of total revenue in fiscal 2025; total revenue $56.7B (fiscal 2025, per the FY2026 earnings release)"},"note":"Upper bound, not a reported value: Cisco's 10-K says hyperscaler AI infrastructure was less than 2% of fiscal 2025 revenue; 2% of $56.7B (fiscal 2025 revenue per the FY2026 earnings release, https://newsroom.cisco.com/c/r/newsroom/en/us/a/y2026/m08/cisco-reports-fourth-quarter-earnings.html) is $1.13B, rounded to $1.13B."}},"committedExpansion":{"low":1.325,"central":1.325,"high":1.325,"unit":"ratio","basis":"Hyperscaler AI orders taken in FY2026 in excess of AI revenue delivered in FY2026 ÷ FY2026 AI revenue","period":"FY2026 (to 25 July 2026)","periodEnd":"2026-07-25","source":{"url":"https://newsroom.cisco.com/c/r/newsroom/en/us/a/y2026/m08/cisco-reports-fourth-quarter-earnings.html","title":"Cisco reports fourth quarter and fiscal year 2026 earnings","publisher":"Cisco","publishedDate":"2026-08-12","accessedDate":"2026-10-22","tier":"T2","locator":"Release: $4 billion of orders taken in Q4, bringing the total for FY 2026 to $9.3 billion; approximately $4 billion of AI infrastructure revenue delivered in FY 2026"},"note":"Cisco says hyperscaler AI infrastructure orders were $9.3B in fiscal 2026 and it delivered about $4B of revenue; (9.3 - 4.0) / 4.0 = 1.325. This counts only orders taken within the year, so it is a lower bound on the backlog (orders from fiscal 2025 not yet delivered are left out); the release gives no AI-specific backlog. System-level, not chip value."},"screening":{"networking-silicon":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"4299c2708456600f0fd31fa55bc036a28ca0be775fbb55e970586767ca87afb4"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"4299c2708456600f0fd31fa55bc036a28ca0be775fbb55e970586767ca87afb4"}}},"alchip":{"schema":"5lc-index-inputs/1","key":"chips-chip-design","entityId":"alchip","name":"Alchip Technologies","kind":"public","commercialActivityVerified":true,"shares":{"merchant-accelerators":{"participates":false,"evidence":{"url":"https://www.alchip.com/en/Newsroom/Alchip_2025_Q4_financial_results","title":"Alchip Technologies reports Q4 and 2025 FY financial results","publisher":"Alchip Technologies","publishedDate":"2026-03-18","accessedDate":"2026-10-22","tier":"T2","locator":"Release: Alchip is a design and production services company for custom ASICs; it sells no chips of its own design, no processor IP"}},"custom-accelerators":{"low":1.687,"central":1.687,"high":1.687,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.alchip.com/en/Newsroom/Alchip_2025_Q4_financial_results","title":"Alchip Technologies reports Q4 and 2025 FY financial results","publisher":"Alchip Technologies","publishedDate":"2026-03-18","accessedDate":"2026-10-22","tier":"T5","locator":"FY2025 revenue $992M; HPC/AI 83%; North America 78%"},"note":"T5 derivation (full inputs). Alchip 2025 revenue $992M, 83% HPC/AI ($823M), 78% North America; data-centre AI ASIC programmes taken as $0.65B (Amazon about $0.5B, others $0.15B), Alchip's partner half $0.325B. 2025 fell from $1.6B because its top customer's new generation only ramps from Q2 2026. 2025 value about $0.325B. Function total $19.27B (was $17.03B before Amazon's programme was re-derived from chip volume, a recorded method decision; see fixes/a recorded method decision/itemisation-changes-a recorded method decision.md) = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $14.7–20.5B (central $16.5B)). Share = 0.325 / 19.27 = 1.687%. Point value: the engine applies the T5 width. Itemisation and 2024 prior in a local evidence capture."},"networking-silicon":{"participates":false,"evidence":{"url":"https://www.alchip.com/en/Newsroom/Alchip_2025_Q4_financial_results","title":"Alchip Technologies reports Q4 and 2025 FY financial results","publisher":"Alchip Technologies","publishedDate":"2026-03-18","accessedDate":"2026-10-22","tier":"T2","locator":"Release: Alchip is a design and production services company for custom ASICs; it sells no chips of its own design, no processor IP"}},"host-cpus":{"participates":false,"evidence":{"url":"https://www.alchip.com/en/Newsroom/Alchip_2025_Q4_financial_results","title":"Alchip Technologies reports Q4 and 2025 FY financial results","publisher":"Alchip Technologies","publishedDate":"2026-03-18","accessedDate":"2026-10-22","tier":"T2","locator":"Release: Alchip is a design and production services company for custom ASICs; it sells no chips of its own design, no processor IP"}},"processor-ip":{"participates":false,"evidence":{"url":"https://www.alchip.com/en/Newsroom/Alchip_2025_Q4_financial_results","title":"Alchip Technologies reports Q4 and 2025 FY financial results","publisher":"Alchip Technologies","publishedDate":"2026-03-18","accessedDate":"2026-10-22","tier":"T2","locator":"Release: Alchip is a design and production services company for custom ASICs; it sells no chips of its own design, no processor IP"}}},"priorShares":{"custom-accelerators":{"low":3.416,"central":4.516,"high":5.693,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2024","periodEnd":"2024-12-31","source":{"url":"https://markets.financialcontent.com/franklincredit/article/gnwcq-2025-3-28-alchip-technologies-sets-2024-financial-records","title":"Alchip Technologies Sets 2024 Financial Records","publisher":"Alchip Technologies (GlobeNewswire, via FinancialContent)","publishedDate":"2025-03-28","accessedDate":"2026-10-22","tier":"T5","locator":"FY 2024 revenue $1.62 billion, up 65.4%; HPC/AI 93% of revenue"},"note":"T5 derivation of the prior-year share. Alchip 2024 revenue $1.62B, 93% HPC/AI ($1.51B); data-centre AI ASIC programmes taken as $1.19B (79% of HPC/AI, the 2025 ratio; range $0.9-1.5B), Alchip's partner half $0.595B (range $0.45-0.75B). 2024 value about $0.595B. Function total $13.17B (was $11.21B before Amazon's 2024 programme was restated with the same in-house multiple, a recorded method decision) = the itemised 2024 values of every candidate plus residual others, built like the 2025 total; candidates without a reported 2024 value are set at half their 2025 value (an assumption: most grew at least 100%). Share = 0.595 / 13.174 = 4.516%. Value range $0.45-0.75B gives the share range 3.416-5.693%. Itemisation in a local evidence capture."}},"activityGrowth":{"current":{"low":992000000,"central":992000000,"high":992000000,"unit":"USD","basis":"Total revenue (about 83-93% HPC and AI ASICs), calendar year, US dollars","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.alchip.com/en/Newsroom/Alchip_2025_Q4_financial_results","title":"Alchip Technologies reports Q4 and 2025 FY financial results","publisher":"Alchip Technologies","publishedDate":"2026-03-18","accessedDate":"2026-10-22","tier":"T2","locator":"FY 2025 revenue $992 million"},"note":"Alchip's revenue fell 39% in 2025 because its top customer's new generation only ramps from Q2 2026 (company says); HPC/AI was 83% of 2025 and 93% of 2024 revenue."},"prior":{"low":1620000000,"central":1620000000,"high":1620000000,"unit":"USD","basis":"Total revenue (about 83-93% HPC and AI ASICs), calendar year, US dollars","period":"CY2024","periodEnd":"2024-12-31","source":{"url":"https://markets.financialcontent.com/franklincredit/article/gnwcq-2025-3-28-alchip-technologies-sets-2024-financial-records","title":"Alchip Technologies Sets 2024 Financial Records","publisher":"Alchip Technologies (GlobeNewswire, via FinancialContent)","publishedDate":"2025-03-28","accessedDate":"2026-10-22","tier":"T2","locator":"FY 2024 revenue reached a record $1.62 billion"}}},"baselineRevenueUsd":{"low":823400000,"central":823400000,"high":823400000,"unit":"USD","basis":"HPC and AI revenue, calendar year, US dollars","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.alchip.com/en/Newsroom/Alchip_2025_Q4_financial_results","title":"Alchip Technologies reports Q4 and 2025 FY financial results","publisher":"Alchip Technologies","publishedDate":"2026-03-18","accessedDate":"2026-10-22","tier":"T2","locator":"FY 2025 revenue $992 million; HPC accounted for 83% of 2025 revenue"},"note":"83% x $992M = $823M (derived); the 83% is HPC and AI together."},"screening":{"custom-accelerators":{"levelLow":2,"levelHigh":4}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"0ae82720ee7bd7aff548923ac4f7992f7667a619b394f547345d332966892eba"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"0ae82720ee7bd7aff548923ac4f7992f7667a619b394f547345d332966892eba"}}},"cerebras":{"schema":"5lc-index-inputs/1","key":"chips-chip-design","entityId":"cerebras","name":"Cerebras","kind":"public","commercialActivityVerified":true,"shares":{"merchant-accelerators":{"low":0.204,"central":0.204,"high":0.204,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.sec.gov/Archives/edgar/data/2021728/000162828026035214/cerebras-424b4.htm","title":"Cerebras Systems Inc. prospectus (Form 424B4)","publisher":"Cerebras Systems / U.S. SEC","publishedDate":"2026-05-14","accessedDate":"2026-10-22","tier":"T5","locator":"Consolidated statement of operations: Hardware $358,440 thousand (2025)"},"note":"T5 derivation (full inputs). Cerebras 2025 hardware revenue $358.44M; its cloud and other services revenue ($151.6M) is service revenue, not chips supplied, and is left out. 2025 value about $0.3584B. Function total $175.74B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $168.7–183.3B (central $175.8B)). Share = 0.3584 / 175.74 = 0.204%. Point value: the engine applies the T5 width. Itemisation and 2024 prior in a local evidence capture."},"custom-accelerators":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/2021728/000162828026035214/cerebras-424b4.htm","title":"Cerebras Systems Inc. prospectus (Form 424B4)","publisher":"Cerebras Systems / U.S. SEC","publishedDate":"2026-05-14","accessedDate":"2026-10-22","tier":"T1","locator":"Business: Cerebras sells its own wafer-scale AI systems and cloud services; no custom chips for other owners"}},"networking-silicon":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/2021728/000162828026035214/cerebras-424b4.htm","title":"Cerebras Systems Inc. prospectus (Form 424B4)","publisher":"Cerebras Systems / U.S. SEC","publishedDate":"2026-05-14","accessedDate":"2026-10-22","tier":"T1","locator":"Business: no networking chip products"}},"host-cpus":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/2021728/000162828026035214/cerebras-424b4.htm","title":"Cerebras Systems Inc. prospectus (Form 424B4)","publisher":"Cerebras Systems / U.S. SEC","publishedDate":"2026-05-14","accessedDate":"2026-10-22","tier":"T1","locator":"Business: no host processors"}},"processor-ip":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/2021728/000162828026035214/cerebras-424b4.htm","title":"Cerebras Systems Inc. prospectus (Form 424B4)","publisher":"Cerebras Systems / U.S. SEC","publishedDate":"2026-05-14","accessedDate":"2026-10-22","tier":"T1","locator":"Business: no processor IP licensing"}}},"priorShares":{"merchant-accelerators":{"low":0.1893,"central":0.1893,"high":0.1893,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2024","periodEnd":"2024-12-31","source":{"url":"https://www.sec.gov/Archives/edgar/data/2021728/000162828026035214/cerebras-424b4.htm","title":"Cerebras Systems Inc. prospectus (Form 424B4)","publisher":"Cerebras Systems / U.S. SEC","publishedDate":"2026-05-14","accessedDate":"2026-10-22","tier":"T5","locator":"Results of operations: Hardware $211,965 thousand (2024)"},"note":"T5 derivation of the prior-year share. Cerebras 2024 hardware revenue $211.965M (prospectus, 2024 column); cloud and other services revenue ($78.3M) is left out, as in 2025. 2024 value about $0.212B. Function total $111.97B = the itemised 2024 values of every candidate plus residual others, built like the 2025 total; candidates without a reported 2024 value are set at half their 2025 value (an assumption: most grew at least 100%). Share = 0.212 / 111.97 = 0.1893%. Point value: the engine applies the T5 width. Itemisation in a local evidence capture."}},"activityGrowth":{"current":{"low":358440000,"central":358440000,"high":358440000,"unit":"USD","basis":"Hardware revenue (wafer-scale systems), calendar year, US dollars","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.sec.gov/Archives/edgar/data/2021728/000162828026035214/cerebras-424b4.htm","title":"Cerebras Systems Inc. prospectus (Form 424B4)","publisher":"Cerebras Systems / U.S. SEC","publishedDate":"2026-05-14","accessedDate":"2026-10-22","tier":"T1","locator":"Results of operations: Hardware $358,440 thousand"},"note":"Hardware only; cloud and other services revenue is left out, as in the shares."},"prior":{"low":211965000,"central":211965000,"high":211965000,"unit":"USD","basis":"Hardware revenue (wafer-scale systems), calendar year, US dollars","period":"CY2024","periodEnd":"2024-12-31","source":{"url":"https://www.sec.gov/Archives/edgar/data/2021728/000162828026035214/cerebras-424b4.htm","title":"Cerebras Systems Inc. prospectus (Form 424B4)","publisher":"Cerebras Systems / U.S. SEC","publishedDate":"2026-05-14","accessedDate":"2026-10-22","tier":"T1","locator":"Results of operations: Hardware $211,965 thousand"}}},"committedExpansion":{"low":48.2,"central":48.2,"high":48.2,"unit":"ratio","basis":"Remaining performance obligations ÷ 2025 total revenue","period":"RPO at 31 December 2025; revenue CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.sec.gov/Archives/edgar/data/2021728/000162828026035214/cerebras-424b4.htm","title":"Cerebras Systems Inc. prospectus (Form 424B4)","publisher":"Cerebras Systems / U.S. SEC","publishedDate":"2026-05-14","accessedDate":"2026-10-22","tier":"T1","locator":"Notes to the consolidated financial statements, Remaining Performance Obligation: $24.6 billion as of December 31, 2025"},"note":"RPO $24.6B at 31 December 2025 (mostly the OpenAI master relationship agreement; about 15% falls in the first 24 months) ÷ total revenue $509.991M (hardware $358.4M plus cloud and other services $151.6M; the RPO covers both, so total revenue is the matching denominator) = 48.2. Pass-through costs and capacity beyond the first 250MW are excluded from the RPO."},"baselineRevenueUsd":{"low":358440000,"central":358440000,"high":358440000,"unit":"USD","basis":"Hardware revenue, calendar year, US dollars","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.sec.gov/Archives/edgar/data/2021728/000162828026035214/cerebras-424b4.htm","title":"Cerebras Systems Inc. prospectus (Form 424B4)","publisher":"Cerebras Systems / U.S. SEC","publishedDate":"2026-05-14","accessedDate":"2026-10-22","tier":"T1","locator":"Results of operations: Hardware $358,440 thousand"},"note":"Hardware only; cloud and other services ($151.6M) left out."},"screening":{"merchant-accelerators":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"1de1af2f753763898ff837e9c4e574e10ef72a52c0148aae54f3504ddc5866b2"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"1de1af2f753763898ff837e9c4e574e10ef72a52c0148aae54f3504ddc5866b2"}}},"guc":{"schema":"5lc-index-inputs/1","key":"chips-chip-design","entityId":"guc","name":"GUC","kind":"public","commercialActivityVerified":true,"shares":{"merchant-accelerators":{"participates":false,"evidence":{"url":"https://www.guc-asic.com/upload/2026_04_16/8_202604161551145dnuwxrHk5.pdf","title":"Global Unichip Corp. 2025 Annual Report","publisher":"Global Unichip Corp.","publishedDate":"2026-03-23","accessedDate":"2026-10-22","tier":"T1","locator":"Business description (PDF p.44): GUC's lines of business are ASIC and wafer products (NT$25.74B, 75.4% of 2025 revenue), NRE (NT$8.03B, 23.5%) and other (1.1%), plus silicon IP (interface and memory IP such as GLink, UCIe, HBM PHY) and cell libraries; PDF p.46: GUC 'ranks among the top pure ASIC design service companies'; no merchant accelerator of its own design"}},"custom-accelerators":{"low":0.6078,"central":0.6078,"high":0.6078,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.guc-asic.com/upload/2026_01_29/8_20260129161502ryd2n2II41.pdf","title":"Global Unichip Corp. and Subsidiaries: Consolidated Financial Statements for the Years Ended December 31, 2025 and 2024 and Independent Auditors' Report","publisher":"Global Unichip Corp.","publishedDate":"2026-01-29","accessedDate":"2026-10-22","tier":"T5","locator":"Note 17 Net revenue, b. Disaggregation (PDF p.34): application type Cloud NT$24,334,142 thousand of NT$34,140,978 thousand (2025); NT$13,256,339 thousand of NT$25,044,192 thousand (2024)"},"note":"T5 derivation (fix pass). GUC's audited 2025 statements split revenue by application: Cloud NT$24.334B of NT$34.141B (2024: Cloud NT$13.256B of NT$25.044B). GUC says its Cloud category 'currently primarily comes from CPU, AI accelerators, BMC and crypto currency projects' (1Q26 results release, https://www.guc-asic.com/upload/2026_04_30/8_20260430170804zqdqffIba1.pdf), and its 4Q25 release says its 3nm-and-below turnkey revenue (54% of fourth-quarter turnkey revenue) came 'mainly from CPU project' (https://www.guc-asic.com/upload/2026_01_29/8_20260129184435tjqzu9PHp1.pdf); GUC gives no accelerator split. Its 3Q25 statements (https://www.guc-asic.com/upload/2025_10_30/8_202510301652058jg1jaVeT1.pdf, revenue by application) show the older 'AI/ML' class at NT$2.418B for the first nine months of 2025 (11% of sales; NT$1.968B, 23%, in Q3) and NT$2.914B for the nine months of 2024 (15%). Judgment: AI accelerators are 30% of Cloud revenue (range 10-50%), which is 21.4% of 2025 sales and 15.9% of 2024 sales. Cloud NT$24.334B at 31.1663 TWD per USD (Federal Reserve G.5A (release of 5 January 2026, https://www.federalreserve.gov/releases/g5a/current/default.htm): 2025 average 31.1663 TWD per USD, 2024 average 32.1064) is $0.781B; 30% is a $0.234B programme value; GUC's partner half is $0.1171B. This replaces the earlier $0.2B (37.5% of sales), which no opened source supported. 2025 value about $0.1171B. Function total $19.27B (was $17.03B before Amazon's programme was re-derived from chip volume, a recorded method decision; see fixes/a recorded method decision/itemisation-changes-a recorded method decision.md) = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $14.7–20.5B (central $16.5B)). Share = 0.1171 / 19.27 = 0.6078%. Point value: the engine applies the T5 width. Itemisation and 2024 prior in a local evidence capture (adjustments made in the fix pass: full-inputs/chips-chip-design/fixes/itemisation-changes.md)."},"networking-silicon":{"participates":false,"evidence":{"url":"https://www.guc-asic.com/upload/2026_04_16/8_202604161551145dnuwxrHk5.pdf","title":"Global Unichip Corp. 2025 Annual Report","publisher":"Global Unichip Corp.","publishedDate":"2026-03-23","accessedDate":"2026-10-22","tier":"T1","locator":"Business description (PDF p.44): GUC's lines of business are ASIC and wafer products (NT$25.74B, 75.4% of 2025 revenue), NRE (NT$8.03B, 23.5%) and other (1.1%), plus silicon IP (interface and memory IP such as GLink, UCIe, HBM PHY) and cell libraries; PDF p.46: GUC 'ranks among the top pure ASIC design service companies'; no networking chip of its own design"}},"host-cpus":{"participates":false,"evidence":{"url":"https://www.guc-asic.com/upload/2026_04_16/8_202604161551145dnuwxrHk5.pdf","title":"Global Unichip Corp. 2025 Annual Report","publisher":"Global Unichip Corp.","publishedDate":"2026-03-23","accessedDate":"2026-10-22","tier":"T1","locator":"Business description (PDF p.44): GUC's lines of business are ASIC and wafer products (NT$25.74B, 75.4% of 2025 revenue), NRE (NT$8.03B, 23.5%) and other (1.1%), plus silicon IP (interface and memory IP such as GLink, UCIe, HBM PHY) and cell libraries; PDF p.46: GUC 'ranks among the top pure ASIC design service companies'; no processor of its own design"}},"processor-ip":{"participates":false,"evidence":{"url":"https://www.guc-asic.com/upload/2026_04_16/8_202604161551145dnuwxrHk5.pdf","title":"Global Unichip Corp. 2025 Annual Report","publisher":"Global Unichip Corp.","publishedDate":"2026-03-23","accessedDate":"2026-10-22","tier":"T1","locator":"Business description (PDF p.44): GUC's lines of business are ASIC and wafer products (NT$25.74B, 75.4% of 2025 revenue), NRE (NT$8.03B, 23.5%) and other (1.1%), plus silicon IP (interface and memory IP such as GLink, UCIe, HBM PHY) and cell libraries; PDF p.46: GUC 'ranks among the top pure ASIC design service companies'; no processor-core IP (its IP portfolio is interface, memory-PHY and mixed-signal IP)"}}},"priorShares":{"custom-accelerators":{"low":0.4699,"central":0.4699,"high":0.4699,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2024","periodEnd":"2024-12-31","source":{"url":"https://www.guc-asic.com/upload/2026_01_29/8_20260129161502ryd2n2II41.pdf","title":"Global Unichip Corp. and Subsidiaries: Consolidated Financial Statements for the Years Ended December 31, 2025 and 2024 and Independent Auditors' Report","publisher":"Global Unichip Corp.","publishedDate":"2026-01-29","accessedDate":"2026-10-22","tier":"T5","locator":"Note 17 Net revenue, b. Disaggregation (PDF p.34): application type Cloud NT$13,256,339 thousand of NT$25,044,192 thousand (2024)"},"note":"T5 derivation of the prior-year share, same method as 2025. Cloud NT$13.256B at 32.1064 TWD per USD (Federal Reserve G.5A (release of 5 January 2026, https://www.federalreserve.gov/releases/g5a/current/default.htm): 2025 average 31.1663 TWD per USD, 2024 average 32.1064) is $0.413B; 30% is a $0.124B programme value; GUC's partner half is $0.0619B. 2024 value about $0.0619B. Function total $13.17B (was $11.21B before Amazon's 2024 programme was restated with the same in-house multiple, a recorded method decision) = the itemised 2024 values of every candidate plus residual others, built like the 2025 total; candidates without a reported 2024 value are set at half their 2025 value (an assumption: most grew at least 100%). Share = 0.0619 / 13.174 = 0.4699%. Point value: the engine applies the T5 width. Itemisation and 2024 prior in a local evidence capture (adjustments made in the fix pass: full-inputs/chips-chip-design/fixes/itemisation-changes.md)."}},"activityGrowth":{"current":{"low":34140978000,"central":34140978000,"high":34140978000,"unit":"TWD","basis":"Total net sales (turnkey, NRE and other), calendar year, New Taiwan dollars, audited","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.guc-asic.com/upload/2026_01_29/8_20260129161502ryd2n2II41.pdf","title":"Global Unichip Corp. and Subsidiaries: Consolidated Financial Statements for the Years Ended December 31, 2025 and 2024 and Independent Auditors' Report","publisher":"Global Unichip Corp.","publishedDate":"2026-01-29","accessedDate":"2026-10-22","tier":"T1","locator":"Note 17 Net revenue (PDF p.33): NT$34,140,978 thousand"},"note":"Total net sales, because GUC reports no AI split; audited."},"prior":{"low":25044192000,"central":25044192000,"high":25044192000,"unit":"TWD","basis":"Total net sales (turnkey, NRE and other), calendar year, New Taiwan dollars, audited","period":"CY2024","periodEnd":"2024-12-31","source":{"url":"https://www.guc-asic.com/upload/2026_01_29/8_20260129161502ryd2n2II41.pdf","title":"Global Unichip Corp. and Subsidiaries: Consolidated Financial Statements for the Years Ended December 31, 2025 and 2024 and Independent Auditors' Report","publisher":"Global Unichip Corp.","publishedDate":"2026-01-29","accessedDate":"2026-10-22","tier":"T1","locator":"Note 17 Net revenue (PDF p.33): NT$25,044,192 thousand"}}},"baselineRevenueUsd":{"low":1095445337,"central":1095445337,"high":1095445337,"unit":"USD","basis":"Total net sales, calendar year, US dollars","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.guc-asic.com/upload/2026_01_29/8_20260129161502ryd2n2II41.pdf","title":"Global Unichip Corp. and Subsidiaries: Consolidated Financial Statements for the Years Ended December 31, 2025 and 2024 and Independent Auditors' Report","publisher":"Global Unichip Corp.","publishedDate":"2026-01-29","accessedDate":"2026-10-22","tier":"T1","locator":"Note 17 Net revenue (PDF p.33): NT$34,140,978 thousand"},"note":"NT$34.141B at 31.1663 TWD per USD (Federal Reserve G.5A (release of 5 January 2026, https://www.federalreserve.gov/releases/g5a/current/default.htm): 2025 average 31.1663 TWD per USD, 2024 average 32.1064); audited; GUC reports no AI split."},"screening":{"custom-accelerators":{"levelLow":2,"levelHigh":4}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"580c2c089a748dd485fdce284cdf76ece6da8052dac5f2ebc9b4e2b7a798de0b"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"580c2c089a748dd485fdce284cdf76ece6da8052dac5f2ebc9b4e2b7a798de0b"}}},"groq":{"schema":"5lc-index-inputs/1","key":"chips-chip-design","entityId":"groq","name":"Groq","kind":"private","commercialActivityVerified":true,"shares":{"merchant-accelerators":{"low":0.01707,"central":0.1423,"high":0.2617,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://eu.36kr.com/en/p/3400923515619718","title":"Groq cut its 2025 revenue forecast to $500 million (The Information, via Xin Dongxi)","publisher":"36Kr (republishing Xin Dongxi, citing The Information)","publishedDate":"2025-07-30","accessedDate":"2026-10-22","tier":"T5","locator":"Groq cut its 2025 revenue forecast from over $2B to about $500M; cloud about $40M"},"note":"T5 derivation (full inputs). Groq's July 2025 investor documents (The Information, via 36Kr) cut its 2025 revenue forecast to about $500M, mostly hardware, with cloud revenue about $40M, so the forecast hardware part is about $0.46B. This is a forecast, not reported revenue: part of the revenue moved to 2026, NVIDIA licensed Groq's technology and hired its leaders in December 2025, and actual 2025 results are not public. It is estimated at $0.25B with a wide range of $0.03-0.46B and flagged off-basis, so the engine widens it further. 2025 value about $0.25B. Function total $175.74B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $168.7–183.3B (central $175.8B)). Share = 0.25 / 175.74 = 0.1423%. Value range $0.03-0.46B gives the share range 0.01707-0.2617%. Itemisation and 2024 prior in a local evidence capture.","offBasis":true},"custom-accelerators":{"participates":false,"evidence":{"url":"https://eu.36kr.com/en/p/3400923515619718","title":"Groq cut its 2025 revenue forecast to $500 million (The Information, via Xin Dongxi)","publisher":"36Kr (republishing Xin Dongxi, citing The Information)","publishedDate":"2025-07-30","accessedDate":"2026-10-22","tier":"T3","locator":"Article: Groq sells LPU hardware and GroqCloud inference; no custom design services, networking chips, host processors or processor IP"}},"networking-silicon":{"participates":false,"evidence":{"url":"https://eu.36kr.com/en/p/3400923515619718","title":"Groq cut its 2025 revenue forecast to $500 million (The Information, via Xin Dongxi)","publisher":"36Kr (republishing Xin Dongxi, citing The Information)","publishedDate":"2025-07-30","accessedDate":"2026-10-22","tier":"T3","locator":"Article: Groq sells LPU hardware and GroqCloud inference; no custom design services, networking chips, host processors or processor IP"}},"host-cpus":{"participates":false,"evidence":{"url":"https://eu.36kr.com/en/p/3400923515619718","title":"Groq cut its 2025 revenue forecast to $500 million (The Information, via Xin Dongxi)","publisher":"36Kr (republishing Xin Dongxi, citing The Information)","publishedDate":"2025-07-30","accessedDate":"2026-10-22","tier":"T3","locator":"Article: Groq sells LPU hardware and GroqCloud inference; no custom design services, networking chips, host processors or processor IP"}},"processor-ip":{"participates":false,"evidence":{"url":"https://eu.36kr.com/en/p/3400923515619718","title":"Groq cut its 2025 revenue forecast to $500 million (The Information, via Xin Dongxi)","publisher":"36Kr (republishing Xin Dongxi, citing The Information)","publishedDate":"2025-07-30","accessedDate":"2026-10-22","tier":"T3","locator":"Article: Groq sells LPU hardware and GroqCloud inference; no custom design services, networking chips, host processors or processor IP"}}},"screening":{"merchant-accelerators":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"73b24424d472782ad01c2393699cef929a12dd96858d1768bdb3a92826e5e1c6"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"73b24424d472782ad01c2393699cef929a12dd96858d1768bdb3a92826e5e1c6"}}},"microsoft":{"schema":"5lc-index-inputs/1","key":"chips-chip-design","entityId":"microsoft","name":"Microsoft","kind":"public","commercialActivityVerified":true,"shares":{"merchant-accelerators":{"participates":false,"evidence":{"url":"https://news.microsoft.com/source/emea/features/maia-200-microsoft-ai-accelerator-azure-2/","title":"Microsoft introduces Maia 200, the next generation of AI technology","publisher":"Microsoft","publishedDate":"2026-01-28","accessedDate":"2026-10-22","tier":"T2","locator":"Article: Maia 200 runs in Azure for Microsoft's and OpenAI's models; Microsoft does not sell it"}},"custom-accelerators":{"low":0.2595,"central":0.7786,"high":1.557,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://news.microsoft.com/source/emea/features/maia-200-microsoft-ai-accelerator-azure-2/","title":"Microsoft introduces Maia 200, the next generation of AI technology","publisher":"Microsoft","publishedDate":"2026-01-28","accessedDate":"2026-10-22","tier":"T5","locator":"Maia 200 deployment has already begun in select U.S. data centers"},"note":"T5 derivation (full inputs). Maia 100 was deployed on a small scale in 2025 and Maia 200 from January 2026; programme value paid to design partners (GUC and Marvell) estimated at $0.1–0.6B (central $0.3B), Microsoft's owner half $0.05–0.3B. 2025 value about $0.15B. Function total $19.27B (was $17.03B before Amazon's programme was re-derived from chip volume, a recorded method decision; see fixes/a recorded method decision/itemisation-changes-a recorded method decision.md) = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $14.7–20.5B (central $16.5B)). Share = 0.15 / 19.27 = 0.7786%. Value range $0.05-0.3B gives the share range 0.2595-1.557%. Volume cross-check (a recorded method decision): no opened source gives Maia unit volumes. Microsoft's Maia 200 post (26 January 2026, https://blogs.microsoft.com/blog/2026/01/26/maia-200-the-ai-accelerator-built-for-inference/) says Maia 200 is deployed in the US Central region with US West 3 next and gives no chip counts, so the partner-revenue-based value (owner half $0.15B) is unchanged. Itemisation and 2024 prior in a local evidence capture."},"networking-silicon":{"low":0.02153,"central":0.1077,"high":0.323,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://news.microsoft.com/source/emea/features/maia-200-microsoft-ai-accelerator-azure-2/","title":"Microsoft introduces Maia 200, the next generation of AI technology","publisher":"Microsoft","publishedDate":"2026-01-28","accessedDate":"2026-10-22","tier":"T5","locator":"Microsoft's in-house silicon"},"note":"T5 derivation (full inputs). Azure Boost DPU and MANA NIC chips in AI servers, estimated at $0.01–0.15B. 2025 value about $0.05B. Function total $46.44B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $35.1–47.1B (central $44.8B)). Share = 0.05 / 46.44 = 0.1077%. Value range $0.01-0.15B gives the share range 0.02153-0.323%. Itemisation and 2024 prior in a local evidence capture."},"host-cpus":{"low":0.09971,"central":0.4986,"high":1.994,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://news.microsoft.com/source/emea/features/maia-200-microsoft-ai-accelerator-azure-2/","title":"Microsoft introduces Maia 200, the next generation of AI technology","publisher":"Microsoft","publishedDate":"2026-01-28","accessedDate":"2026-10-22","tier":"T5","locator":"Microsoft's in-house silicon"},"note":"T5 derivation (full inputs). Cobalt Arm processors used as AI-server hosts, estimated at $0.01–0.2B. 2025 value about $0.05B. Function total $10.03B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $7.0–12.9B (central $9.6B)). Share = 0.05 / 10.03 = 0.4986%. Value range $0.01-0.2B gives the share range 0.09971-1.994%. Itemisation and 2024 prior in a local evidence capture."},"processor-ip":{"participates":false,"evidence":{"url":"https://news.microsoft.com/source/emea/features/maia-200-microsoft-ai-accelerator-azure-2/","title":"Microsoft introduces Maia 200, the next generation of AI technology","publisher":"Microsoft","publishedDate":"2026-01-28","accessedDate":"2026-10-22","tier":"T2","locator":"Article: Microsoft does not license processor cores"}}},"screening":{"custom-accelerators":{"levelLow":1,"levelHigh":3},"networking-silicon":{"levelLow":1,"levelHigh":2},"host-cpus":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"21f7fb478dca181ac3f10e4c353659dfe900f1083d8f98c86442f7751b5a6183"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"21f7fb478dca181ac3f10e4c353659dfe900f1083d8f98c86442f7751b5a6183"}}},"tencent":{"schema":"5lc-index-inputs/1","key":"chips-chip-design","entityId":"tencent","name":"Tencent","kind":"public","commercialActivityVerified":false,"shares":{"merchant-accelerators":{"participates":false,"evidence":{"url":"https://eu.36kr.com/en/p/3972785267208455","title":"Tencent's Zixiao chips and its in-house computing plan","publisher":"36Kr (authorised republication of Blue Hole Business)","publishedDate":"2026-09-07","accessedDate":"2026-10-22","tier":"T3","locator":"Article: Zixiao chips serve Tencent's internal businesses; Tencent does not sell accelerators"}},"custom-accelerators":{"low":0.15,"central":0.3,"high":0.7501,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://eu.36kr.com/en/p/3972785267208455","title":"Tencent's Zixiao chips and its in-house computing plan","publisher":"36Kr (authorised republication of Blue Hole Business)","publishedDate":"2026-09-07","accessedDate":"2026-10-22","tier":"T5","locator":"Zixiao 2.0 jointly developed with Enflame; internal businesses led deployment"},"note":"T5 derivation (full inputs). Tencent's Zixiao 2.0 inference chips, jointly developed with Enflame: Enflame's 2025 revenue linked to Tencent (83.79% of CNY 990.2M, $0.1156B at 7.18) is taken as the programme value, split equally between Tencent as owner and Enflame as design partner (range $0.0289-0.1445B for each half). Tencent targets 30% of new computing power and 100,000 cards by end-2026; 2025 deployments were internal. 2025 value about $0.0578B. Function total $19.27B (was $17.03B before Amazon's programme was re-derived from chip volume, a recorded method decision; see fixes/a recorded method decision/itemisation-changes-a recorded method decision.md) = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $14.7–20.5B (central $16.5B)). Share = 0.0578 / 19.27 = 0.3%. Value range $0.0289-0.144B gives the share range 0.15-0.7501%. Itemisation and 2024 prior in a local evidence capture."},"networking-silicon":{"low":0,"central":0.04307,"high":0.1292,"unit":"percent","basis":"Share of the function's 2025 value in US dollars (fiscal year closest to calendar 2025)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://eu.36kr.com/en/p/3972785267208455","title":"Tencent's Zixiao chips and its in-house computing plan","publisher":"36Kr (authorised republication of Blue Hole Business)","publishedDate":"2026-09-07","accessedDate":"2026-10-22","tier":"T5","locator":"Penglai Lab in-house chips (Zixiao, Canghai, Xuanling NIC)"},"note":"T5 derivation (full inputs). Xuanling smart NICs in Tencent's AI servers, estimated at $0–0.06B. 2025 value about $0.02B. Function total $46.44B = the itemised 2025 values of every candidate plus residual others (the same itemisation for every company; the card's own total is $35.1–47.1B (central $44.8B)). Share = 0.02 / 46.44 = 0.04307%. Value range $0-0.06B gives the share range 0-0.1292%. Itemisation and 2024 prior in a local evidence capture."},"host-cpus":{"participates":false,"evidence":{"url":"https://eu.36kr.com/en/p/3972785267208455","title":"Tencent's Zixiao chips and its in-house computing plan","publisher":"36Kr (authorised republication of Blue Hole Business)","publishedDate":"2026-09-07","accessedDate":"2026-10-22","tier":"T3","locator":"Article: Tencent's in-house chips are AI inference, video transcoding and NIC chips; no host processor"}},"processor-ip":{"participates":false,"evidence":{"url":"https://eu.36kr.com/en/p/3972785267208455","title":"Tencent's Zixiao chips and its in-house computing plan","publisher":"36Kr (authorised republication of Blue Hole Business)","publishedDate":"2026-09-07","accessedDate":"2026-10-22","tier":"T3","locator":"Article: no processor-core licensing"}}},"screening":{"custom-accelerators":{"levelLow":1,"levelHigh":3},"networking-silicon":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"56deb8d71555a07dafe69df1eadd78d64986ec0575aadec9282c5e73dcaf09a5"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"56deb8d71555a07dafe69df1eadd78d64986ec0575aadec9282c5e73dcaf09a5"}}}},"ratings":{"nvidia":{"a":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"nvidia","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"merchant-accelerators":{"alternatives":{"level":4,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","date":"2026-02-25","support":"NVIDIA names AMD, Huawei and Intel as accelerator rivals (Item 1, Competition); only AMD supplies comparable GPUs at scale outside China. Full date shown on the source."},{"url":"https://www.constellationr.com/insights/news/intel-defends-gaudi-3-it-misses-2024-sales-targets","date":"2024-11-01","support":"Intel said it would not reach its $500 million Gaudi revenue target for 2024, so Gaudi is not a parity alternative. Full date shown on the source."},{"url":"https://ir.amd.com/news-events/press-releases/detail/1260/amd-and-openai-announce-strategic-partnership-to-deploy-6-gigawatts-of-amd-gpus","date":"2025-10-06","support":"AMD says its OpenAI work began with MI300X: one qualified but partial merchant alternative. Full date shown on the source."}]},"switchingTime":{"level":4,"evidence":[{"url":"https://ir.amd.com/news-events/press-releases/detail/1260/amd-and-openai-announce-strategic-partnership-to-deploy-6-gigawatts-of-amd-gpus","date":"2025-10-06","support":"AMD says OpenAI's first 1 GW of MI450 begins in the second half of 2026, a year after signing; 6 GW is multi-year. Full date shown on the source."},{"url":"https://newsroom.amd.com/news/amd-and-meta-announce-expanded-strategic-partnersh/","date":"2026-02-24","support":"AMD says shipments for Meta's first gigawatt begin in H2 2026; the 6 GW deal spans multiple generations. Full date shown on the source."}]},"capabilityGap":{"level":5,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","date":"2026-02-25","support":"Data Center Compute revenue was $162,361 million in fiscal 2026 (notes, revenue by specialized markets). Full date shown on the source."},{"url":"https://www.sec.gov/Archives/edgar/data/2488/000000248826000018/amd-20251227.htm","date":"2026-02-04","support":"AMD's whole Data Center segment, server CPUs included, was $16,635 million in 2025 (MD&A, segment results): no alternative at scale. Full date shown on the source."}]}},"networking-silicon":{"alternatives":{"level":3,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","date":"2026-02-25","support":"Networking rivals named include Broadcom, Cisco, Marvell and Arista (Item 1, Competition); none is named for NVLink scale-up. Full date shown on the source."},{"url":"https://www.delloro.com/news/ai-back-end-networks-continue-their-shift-to-ethernet-now-accounting-for-over-two-thirds-of-3q-2025-switch-sales-in-ai-clusters/","date":"2025-12-05","support":"Ethernet is over two-thirds of AI back-end switch sales; Accton, Arista and Cisco gained the most share. Full date shown on the source."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://www.theregister.com/2026/04/07/ualink_2_specs/","date":"2026-04-07","support":"UALink chair says 1.0 chips reach labs in H2 2026 and products later in 2027, so scale-up switching takes 1-2 years. Full date shown on the source."},{"url":"https://www.delloro.com/news/ai-back-end-networks-continue-their-shift-to-ethernet-now-accounting-for-over-two-thirds-of-3q-2025-switch-sales-in-ai-clusters/","date":"2025-12-05","support":"Scale-out moved toward Ethernet over about a year: Ethernet share rose from under half to over two-thirds. Full date shown on the source."}]},"capabilityGap":{"level":4,"evidence":[{"url":"https://www.theregister.com/2026/04/07/ualink_2_specs/","date":"2026-04-07","support":"Only UALink 3.0, about a year after April 2026, reaches parity with NVLink in performance and cadence. Full date shown on the source."},{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","date":"2026-02-25","support":"Blackwell racks connect 36 Grace CPUs and 72 GPUs over NVLink (Item 1, Our Company); no open alternative ships at that scale yet. Full date shown on the source."}]}},"host-cpus":{"alternatives":{"level":3,"evidence":[{"url":"https://www.intel.com/content/www/us/en/newsroom/news/artificial-intelligence/new-intel-xeon-6-cpus-maximize-gpu-ai-performance.html","date":"2025-05-22","support":"Intel says a Xeon 6 processor is the host CPU in NVIDIA DGX B300, so x86 hosts are qualified for NVIDIA GPUs. Full date shown on the source."},{"url":"https://www.theregister.com/2026/02/13/amd_intel_market_share/","date":"2026-02-13","support":"AMD held 28.8% and Intel just over 71% of server processors in Q4 2025 (Mercury Research): two x86 alternatives. Full date shown on the source."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","date":"2026-02-25","support":"GB-class racks are built as 36 Grace CPUs with 72 Blackwell GPUs (Item 1), so swapping hosts means a different system design. Full date shown on the source."},{"url":"https://www.intel.com/content/www/us/en/newsroom/news/artificial-intelligence/intel-and-nvidia-to-jointly-develop-ai-infrastructure-and-personal-computing-products.html","date":"2025-09-18","support":"NVLink-connected NVIDIA-custom x86 CPUs from Intel were only announced in September 2025; none ships yet. Full date shown on the source."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://www.intel.com/content/www/us/en/newsroom/news/artificial-intelligence/new-intel-xeon-6-cpus-maximize-gpu-ai-performance.html","date":"2025-05-22","support":"x86 hosts serve DGX B300 nodes, not the NVLink-coherent 72-GPU rack; a material gap for rack-scale systems (inference). Full date shown on the source."},{"url":"https://www.intel.com/content/www/us/en/newsroom/news/artificial-intelligence/intel-and-nvidia-to-jointly-develop-ai-infrastructure-and-personal-computing-products.html","date":"2025-09-18","support":"Intel and NVIDIA will connect their architectures with NVLink only in future custom parts. Full date shown on the source."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"nvidia","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"merchant-accelerators":{"alternatives":{"level":3,"evidence":[{"url":"https://www.anthropic.com/news/expanding-our-use-of-google-cloud-tpus-and-services","date":"2025-10-23","support":"Anthropic says it runs Claude on Google TPUs, Amazon Trainium and NVIDIA GPUs, with up to one million TPUs and well over 1 GW online in 2026 (intro). Date form: full date shown."},{"url":"https://ir.amd.com/news-events/press-releases/detail/1260/amd-and-openai-announce-strategic-partnership-to-deploy-6-gigawatts-of-amd-gpus","date":"2025-10-06","support":"AMD and OpenAI agree up to 6 GW of Instinct GPUs; the first 1 GW of MI450 is set to begin in the second half of 2026. Date form: full date shown."},{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","date":"2026-02-25","support":"NVIDIA's 10-K names AMD, Huawei and Intel as GPU competitors, plus in-house chips at Alibaba, Alphabet, Amazon, Baidu and Microsoft (Competition). Date form: full filing date from the EDGAR index."}]},"switchingTime":{"level":4,"evidence":[{"url":"https://ir.amd.com/news-events/press-releases/detail/1260/amd-and-openai-announce-strategic-partnership-to-deploy-6-gigawatts-of-amd-gpus","date":"2025-10-06","support":"A deal signed in October 2025 reaches its first gigawatt only in H2 2026 and is a multi-year, multi-generation 6 GW plan; replacing NVIDIA's volume takes several such tranches. Date form: full date shown."},{"url":"https://ir.amd.com/news-events/press-releases/detail/1279/","date":"2026-02-24","support":"Meta's up to 6 GW AMD agreement is likewise multi-year and multi-generation, with the first gigawatt shipping from H2 2026. Date form: full date shown."},{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","date":"2026-02-25","support":"NVIDIA says it has paid premiums and signed long-term supply and capacity agreements, and has seen lead times over 12 months (Item 1A, supply risks). Date form: full filing date from the EDGAR index."}]},"capabilityGap":{"level":4,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","date":"2026-02-25","support":"NVIDIA's Compute & Networking segment revenue was $193,479 million in fiscal 2026 (segment note). Date form: full filing date from the EDGAR index."},{"url":"https://www.sec.gov/Archives/edgar/data/2488/000000248826000018/amd-20251227.htm","date":"2026-02-04","support":"AMD's whole Data Center segment, EPYC plus Instinct, was $16,635 million in 2025 (MD&A): a small fraction of NVIDIA's scale. Date form: full filing date from the EDGAR index."},{"url":"https://the-decoder.com/nvidia-sets-new-mlperf-records-with-288-gpus-while-amd-and-intel-focus-on-different-battles/","date":"2026-04-02","support":"In MLPerf Inference v6.0 NVIDIA ran every new model and the largest system (288 GPUs); AMD trailed the B300 in one test and skipped DeepSeek-R1. Date form: full date shown."}]}},"networking-silicon":{"alternatives":{"level":3,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","date":"2026-02-25","support":"NVIDIA's 10-K lists networking competitors AMD, Arista, Broadcom, Cisco, HPE, Huawei, Intel, Lumentum and Marvell (Competition). Date form: full filing date from the EDGAR index."},{"url":"https://www.theregister.com/networks/2026/06/02/marvell-enters-the-ai-network-fray-with-1024-tbps-switch-silicon/5250180","date":"2026-06-02","support":"Broadcom's Tomahawk 6 is already shipping and Cisco's G300 is listed among shipping 102.4T chips; Marvell's T100 only starts sampling this quarter. Date form: full date shown."},{"url":"https://www.theregister.com/2026/04/07/ualink_2_specs/","date":"2026-04-07","support":"For scale-up links, the UALink chair says versions 1.0 and 2.0 \"won't be full competitors to Nvidia\" (NVLink). Date form: full date shown."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://www.theregister.com/2026/04/07/ualink_2_specs/","date":"2026-04-07","support":"UALink 1.0 chips reach labs in H2 2026 and products later in 2027, so no open scale-up alternative to NVLink ships before then. Date form: full date shown."},{"url":"https://www.delloro.com/news/ai-back-end-networks-continue-their-shift-to-ethernet-now-accounting-for-over-two-thirds-of-3q-2025-switch-sales-in-ai-clusters/","date":"2025-12-05","support":"Ethernet rose from under half of AI back-end switch sales a year earlier to over two-thirds in 3Q 2025: scale-out fabrics can be rebalanced in about a year. Date form: full date shown."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://www.theregister.com/2026/04/07/ualink_2_specs/","date":"2026-04-07","support":"The chair expects performance and release-cadence parity with NVIDIA only at UALink version 3.0, about a year after April 2026. Date form: full date shown."},{"url":"https://www.delloro.com/news/ai-back-end-networks-switch-sales-surpass-front-end-networks-for-the-first-time-in-2q2026/","date":"2026-09-03","support":"NVIDIA ranks second in Ethernet AI back-end switch sales, close behind Celestica and ahead of Arista, so scale-out alternatives are near its level. Date form: full date shown."}]}},"host-cpus":{"alternatives":{"level":2,"evidence":[{"url":"https://pcper.com/2025/05/new-intel-xeon-6-cpus-designed-for-nvidia-dgx-b300/","date":"2025-05-26","support":"NVIDIA's own DGX B300 pairs its GPUs with the 64-core Intel Xeon 6776P, so x86 hosts are qualified alternatives to Grace. Date form: full date shown."},{"url":"https://hothardware.com/news/amd-record-x86-market-share-cut-intels-chip-lead","date":"2026-08-21","support":"Mercury Research: AMD holds 34.5% and Intel 65.5% of x86 server CPU share in Q2 2026, so two large x86 suppliers are in volume. Date form: full date shown (reports Mercury Research data)."},{"url":"https://newsroom.arm.com/blog/half-of-compute-shipped-to-top-hyperscalers-in-2025-will-be-arm-based","date":"2025-04-01","support":"Arm says Graviton, Axion, Cobalt and NVIDIA Grace are all Neoverse-based, so other Arm host chips exist (captive to their makers). Date form: full date shown."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://pcper.com/2025/05/new-intel-xeon-6-cpus-designed-for-nvidia-dgx-b300/","date":"2025-05-26","support":"Because DGX B300 already ships with Xeon hosts, moving Grace volume to an x86 host is a platform choice, not a requalification of the host CPU. Date form: full date shown."},{"url":"https://ir.amd.com/news-events/press-releases/detail/1295/amd-reports-second-quarter-2026-financial-results","date":"2026-08-04","support":"AMD says EPYC demand is accelerating and 6th Gen EPYC server CPUs have launched (Q2 2026 results). Date form: full date shown."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://investor.nvidia.com/news/press-release-details/2026/Meta-Builds-AI-Infrastructure-With-NVIDIA/","date":"2026-02-17","support":"NVIDIA says Meta's 'Grace-only' deployment is aimed at performance per watt, so Grace's edge is efficiency rather than a capability x86 hosts lack. Date form: full date shown (NVIDIA release)."},{"url":"https://pcper.com/2025/05/new-intel-xeon-6-cpus-designed-for-nvidia-dgx-b300/","date":"2025-05-26","support":"The DGX B300 runs its Blackwell Ultra GPUs with an x86 Xeon host, showing x86 hosts are workable with NVIDIA's best accelerators. Date form: full date shown."}]}}}}},"broadcom":{"a":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"broadcom","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"custom-accelerators":{"alternatives":{"level":3,"evidence":[{"url":"https://www.techpartner.news/news/google-preparing-to-partner-with-mediatek-on-next-ai-chip-615763","date":"2025-03-18","support":"Google is preparing to add MediaTek for its next TPU; Broadcom had been its exclusive partner (Reuters). Full date shown on the source."},{"url":"https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm","date":"2026-03-11","support":"Marvell names ASIC design firms such as Alchip and GUC as competitors (Item 1, Competition). Full date shown on the source."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://www.techpartner.news/news/google-preparing-to-partner-with-mediatek-on-next-ai-chip-615763","date":"2025-03-18","support":"The MediaTek-designed TPUs are to be made the following year, about 1-2 years after engagement (Reuters). Full date shown on the source."},{"url":"https://www.taipeitimes.com/News/biz/archives/2025/05/30/2003837724","date":"2025-05-30","support":"Alchip expects small-volume production of AWS's 3nm Trainium 3 at end-2025 and a ramp in 2026. Full date shown on the source."}]},"capabilityGap":{"level":4,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1730168/000173016825000116/avgo-11022025x8kxex99.htm","date":"2025-12-11","support":"Broadcom guided Q1 fiscal 2026 AI revenue to $8.2 billion, driven by custom AI accelerators (company says). Full date shown on the source."},{"url":"https://www.taipeitimes.com/News/biz/archives/2025/05/30/2003837724","date":"2025-05-30","support":"Alchip, a leading rival, was only starting small-volume 3nm production at end-2025: a large scale gap. Full date shown on the source."}]}},"networking-silicon":{"alternatives":{"level":3,"evidence":[{"url":"https://investor.nvidia.com/news/press-release-details/2025/NVIDIA-Spectrum-X-Ethernet-Switches-Speed-Up-Networks-for-Meta-and-Oracle/default.aspx","date":"2025-10-13","support":"Meta will integrate NVIDIA Spectrum Ethernet switches into FBOSS and Oracle builds on Spectrum-X. Full date shown on the source."},{"url":"https://blogs.cisco.com/news/cisco-silicon-one-g200-ai-ml-chip-powers-new-systems-for-hyperscalers-and-enterprises","date":"2024-10-16","support":"Meta planned to deploy the Cisco 8501 built on Cisco Silicon One G200 (Cisco blog). Full date shown on the source."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://blogs.cisco.com/news/cisco-silicon-one-g200-ai-ml-chip-powers-new-systems-for-hyperscalers-and-enterprises","date":"2024-10-16","support":"Meta announced Cisco G200 switches as a plan in October 2024; new switch silicon takes about a year to deploy. Full date shown on the source."},{"url":"https://www.sec.gov/Archives/edgar/data/1807794/000162828026043303/crdo-20260502.htm","date":"2026-06-15","support":"Credo says customer evaluation of networking chips may take several months or more than a year (Item 1A). Full date shown on the source."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://www.storagereview.com/news/broadcom-ships-tomahawk-6-switch-series-unleashing-102-4-tbps-switching-capacity","date":"2025-06-09","support":"Tomahawk 6 offers 102.4 Tbps, doubling the bandwidth of existing Ethernet switches when it shipped. Full date shown on the source."},{"url":"https://www.sec.gov/Archives/edgar/data/1730168/000173016825000116/avgo-11022025x8kxex99.htm","date":"2025-12-11","support":"Broadcom says its AI growth is driven by custom accelerators and Ethernet AI switches; rivals trail at scale. Full date shown on the source."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"broadcom","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"custom-accelerators":{"alternatives":{"level":3,"evidence":[{"url":"https://counterpointresearch.com/de/insights/AI-Server-Compute-ASIC-Shipments-to-Triple-by-2027","date":"2026-01","support":"Counterpoint: Broadcom about 60% of design-partner share in 2027, Marvell about 8%; Alchip and a Google-MediaTek alliance are the other qualified partners. Date form: month only, because the page shows 26 January but its dateline year is inconsistent."},{"url":"https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm","date":"2026-03-11","support":"Marvell's 10-K lists Alchip, Global Unichip, MediaTek and Socionext among its direct competitors (Competition). Date form: full filing date from the EDGAR index."},{"url":"https://technode.com/2025/03/19/google-partners-with-mediatek-for-next-gen-tpu-shifting-orders-from-broadcom/","date":"2025-03-19","support":"The Information (via TechNode): Google gave MediaTek a share of TPU design orders that Broadcom previously handled. Date form: full date shown."}]},"switchingTime":{"level":4,"evidence":[{"url":"https://siliconanalysts.com/analysis/custom-ai-chip-economics-tapeout-speed-vs-cost-2026","date":"2026-08-27","support":"Conventional advanced-node ASIC timeline is 18 to 36 months from architecture freeze to first silicon, then 6 to 12 months of bring-up. Date form: full date shown."},{"url":"https://futurumgroup.com/insights/jalapeo-in-nine-months-did-ai-just-break-chip-design-timelines/","date":"2026-06-26","support":"Concept to tape-out in 18 to 24 months 'is considered fast' for a frontier accelerator. Date form: full date shown."},{"url":"https://www.sdxcentral.com/news/broadcom-secures-deal-with-google-for-next-gen-ai-racks-and-custom-tpus/","date":"2026-04-07","support":"Broadcom agreed to develop and supply Google's custom TPU silicon and holds a supply assurance agreement through 2031; a 3.5 GW Anthropic TPU programme starts in 2027. Date form: full date shown."}]},"capabilityGap":{"level":4,"evidence":[{"url":"https://counterpointresearch.com/de/insights/AI-Server-Compute-ASIC-Shipments-to-Triple-by-2027","date":"2026-01","support":"Counterpoint: MediaTek's TPU v8x high-volume ramp is delayed about a year, and Marvell's design share falls to about 8% in 2027. Date form: month only, because the page shows 26 January but its dateline year is inconsistent."},{"url":"https://www.sdxcentral.com/news/broadcom-slams-ai-silicon-fueled-q3-promises-double-double-on-the-way/","date":"2026-09-03","support":"Broadcom's CEO calls Anthropic its largest XPU customer in 2027 and OpenAI its second largest, and says 'We have secured the supply' for the outlook. Date form: full date shown (trade press on Broadcom's earnings call)."}]}},"networking-silicon":{"alternatives":{"level":3,"evidence":[{"url":"https://www.theregister.com/networks/2026/06/02/marvell-enters-the-ai-network-fray-with-1024-tbps-switch-silicon/5250180","date":"2026-06-02","support":"The Register: Broadcom's Tomahawk 6 has shipped since 2025, Cisco's G300 is listed among shipping 102.4T parts, and Marvell's T100 is only about to sample. Date form: full date shown."},{"url":"https://www.sdxcentral.com/news/ai-hyperscale-and-agentic-demands-helped-cisco-smash-revenue-records/","date":"2026-08-13","support":"Cisco says three hyperscalers now use its Silicon One P200 for scale-across networks, and it took more orders in Q4 (call summary). Date form: full date shown (trade press on Cisco's earnings call)."},{"url":"https://www.delloro.com/news/ai-back-end-networks-switch-sales-surpass-front-end-networks-for-the-first-time-in-2q2026/","date":"2026-09-03","support":"NVIDIA ranks second and Cisco fourth in Ethernet AI back-end switch sales in 2Q 2026, so two other vendors already sell at scale. Date form: full date shown."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://investor.cisco.com/news/news-details/2026/Cisco-Announces-New-Silicon-One-G300-Advanced-Systems-and-Optics-to-Power-and-Scale-AI-Data-Centers-for-the-Agentic-Era/","date":"2026-02-10","support":"Cisco says the 102.4T G300 and its systems 'will ship this year' (February 2026), so other silicon reaches hyperscalers only in 2026-27. Date form: full date shown."},{"url":"https://www.theregister.com/networks/2026/06/02/marvell-enters-the-ai-network-fray-with-1024-tbps-switch-silicon/5250180","date":"2026-06-02","support":"Marvell's competing 102.4T switch only begins sampling in Q2 2026, so a hyperscaler redesign would run over one to two years. Date form: full date shown."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.theregister.com/networks/2026/06/02/marvell-enters-the-ai-network-fray-with-1024-tbps-switch-silicon/5250180","date":"2026-06-02","support":"The Register: rivals already ship equivalent 102.4T products, so the best alternative is close to Broadcom's level. Date form: full date shown."},{"url":"https://www.sdxcentral.com/news/ai-hyperscale-and-agentic-demands-helped-cisco-smash-revenue-records/","date":"2026-08-13","support":"Cisco took $9.3 billion of hyperscaler AI orders in fiscal 2026, about 60% Silicon One systems (company statement on its call). Date form: full date shown (trade press on Cisco's earnings call)."}]}}}}},"huawei":{"a":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"huawei","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"merchant-accelerators":{"alternatives":{"level":3,"evidence":[{"url":"https://kfgo.com/?p=1285523","date":"2026-04-01","support":"IDC (via Reuters): in China in 2025, T-Head shipped about 265,000 cards and Cambricon and Kunlunxin 116,000 each. Full date shown on the source."},{"url":"https://kfgo.com/?p=1285523","date":"2026-04-01","support":"Nvidia still shipped about 2.2 million cards in China in 2025, but U.S. controls cut off its most advanced products. Full date shown on the source."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://www.thestandard.com.hk/innovation/article/311827/China-tells-tech-firms-to-stop-buying-Nvidias-AI-chips-FT-reports","date":"2025-09-17","support":"China's regulator told ByteDance and Alibaba to end testing and orders of Nvidia's RTX Pro 6000D (Reuters). Full date shown on the source."},{"url":"https://kfgo.com/?p=1285523","date":"2026-04-01","support":"Many local projects reportedly carried implicit directives to 'buy Chinese', so Huawei users must port to other domestic stacks. Full date shown on the source."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://kfgo.com/?p=1285523","date":"2026-04-01","support":"Huawei's roughly 812,000 chips were about half of domestic-brand shipments, about three times the next vendor. Full date shown on the source."},{"url":"https://www.huawei.com/en/news/2025/9/hc-xu-keynote-speech","date":"2025-09-18","support":"Huawei says 'we don't have access to advanced process nodes', a limit shared by domestic rivals (keynote dateline 18 September 2025). Full date shown on the source."}]}},"host-cpus":{"alternatives":{"level":2,"evidence":[{"url":"https://www.khaleejtimes.com/business/tech/intel-amd-sign-long-term-server-cpu-deals-with-chinese-clients-as-prices-surge-sources-say","date":"2026-07-23","support":"Intel and AMD still supply server CPUs to Chinese customers under one-year volume commitments (Reuters). Full date shown on the source."},{"url":"https://www.cls.cn/detail/2337024","date":"2026-04-07","support":"Hygon, a domestic x86 alternative, reported 2025 revenue of CNY 14.38 billion (annual report summary). Full date shown on the source."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.khaleejtimes.com/business/tech/intel-amd-sign-long-term-server-cpu-deals-with-chinese-clients-as-prices-surge-sources-say","date":"2026-07-23","support":"Reuters: Intel lead times reached as long as six months for some server CPUs in China. Full date shown on the source."},{"url":"https://www.huawei.com/en/news/2025/9/hc-xu-keynote-speech","date":"2025-09-18","support":"Kunpeng is being tied into SuperPoDs (TaiShan 950 in Q1 2026); ordinary Ascend servers can take other hosts (inference). Full date shown on the source."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.khaleejtimes.com/business/tech/intel-amd-sign-long-term-server-cpu-deals-with-chinese-clients-as-prices-surge-sources-say","date":"2026-07-23","support":"Shortage: some server CPU prices in China rose more than 40% since January 2026, a small capacity gap. Full date shown on the source."},{"url":"https://www.cls.cn/detail/2337024","date":"2026-04-07","support":"Hygon's 2025 revenue grew 56.92%, showing domestic x86 capacity that can host AI servers. Full date shown on the source."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"huawei","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"merchant-accelerators":{"alternatives":{"level":3,"evidence":[{"url":"https://whbl.com/2026/04/01/chinese-chipmakers-claim-nearly-half-of-local-market-as-nvidias-lead-shrinks/","date":"2026-04-01","support":"IDC (via Reuters): in China, Huawei shipped about 812,000 cards in 2025; T-Head about 265,000; Kunlunxin and Cambricon about 116,000 each. Date form: full date shown (Reuters report on IDC data, wire copy)."},{"url":"https://www.nst.com.my/amp/business/corporate/2026/07/1484039/china-plans-let-top-ai-firms-buy-limited-nvidia-h200-chips","date":"2026-07-09","support":"Beijing may allow about 200,000 or fewer NVIDIA H200 chips in total for Alibaba, ByteDance and DeepSeek, less than half requested. Date form: full date shown (Reuters story)."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://gulfnews.com/technology/huawei-to-double-output-of-top-ai-chip-as-nvidia-wavers-in-china-1.500287469","date":"2025-09-30","support":"Bloomberg: Huawei plans about 600,000 910C chips in 2026, roughly double 2025; rivals would need a similar doubling on the same SMIC process. Date form: full date from the page's last-updated line (Bloomberg report)."},{"url":"https://www.theregister.com/systems/2026/05/22/alibaba-just-admitted-its-struggling-to-keep-up-with-rival-chipmakers-and-ai-shops/5244665","date":"2026-05-22","support":"The Register: Beijing has not let local buyers acquire Nvidia chips, so foreign supply is not an available fallback. Date form: full date shown."}]},"capabilityGap":{"level":4,"evidence":[{"url":"https://whbl.com/2026/04/01/chinese-chipmakers-claim-nearly-half-of-local-market-as-nvidias-lead-shrinks/","date":"2026-04-01","support":"Huawei's 812,000 cards were about half of all domestic shipments in 2025; the next domestic vendor shipped about a third of that. Date form: full date shown (Reuters report on IDC data, wire copy)."},{"url":"https://gulfnews.com/technology/huawei-to-double-output-of-top-ai-chip-as-nvidia-wavers-in-china-1.500287469","date":"2025-09-30","support":"Bloomberg: SMIC is stuck at 7 nm, yields are underwhelming and Huawei relies on stockpiled foreign HBM, limiting what any domestic rival can add. Date form: full date from the page's last-updated line (Bloomberg report)."},{"url":"https://www.theregister.com/systems/2026/05/22/alibaba-just-admitted-its-struggling-to-keep-up-with-rival-chipmakers-and-ai-shops/5244665","date":"2026-05-22","support":"The Register: T-Head, the second-largest domestic vendor, can't get much hardware made, and Chinese fabs can't yet match TSMC. Date form: full date shown."}]}},"host-cpus":{"alternatives":{"level":2,"evidence":[{"url":"https://sinolytics.de/global-business-news/blog/technology/china-cpu-self-sufficiency-policy-procurement/","date":"2026-07-06","support":"Sinolytics: Huawei Kunpeng and Hygon are the frontrunners among domestic CPUs; Phytium, Zhaoxin and Loongson are also in the market. Date form: full date shown."},{"url":"https://thenextweb.com/news/china-telecom-server-deal-huawei-kunpeng-1-7bn","date":"2026-06-24","support":"A June 2026 China Telecom tender split server orders: 28,000 Arm servers tied to Kunpeng and 12,000 C86 (x86-compatible) servers. Date form: full date shown."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://thenextweb.com/news/china-telecom-server-deal-huawei-kunpeng-1-7bn","date":"2026-06-24","support":"China Telecom bought Arm (Kunpeng) and C86 servers in the same tender, so buyers already run both instruction sets. Date form: full date shown."},{"url":"https://sinolytics.de/global-business-news/blog/technology/china-cpu-self-sufficiency-policy-procurement/","date":"2026-07-06","support":"Sinolytics: domestic CPUs hold about 20% of China's overall CPU market, so most volume still sits with other suppliers. Date form: full date shown."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://heise.de/-10385441","date":"2025-05-15","support":"Hygon's C86-5G is estimated at around Zen 3 level by heise (roadmap-based, not benchmarked), so the nearest domestic rival is a generation behind at most. Date form: full date shown."},{"url":"https://sinolytics.de/global-business-news/blog/technology/china-cpu-self-sufficiency-policy-procurement/","date":"2026-07-06","support":"Sinolytics says Kunpeng and Hygon benefit from stronger performance; no gap against Intel or AMD is quantified. Date form: full date shown."}]}}}}},"amd":{"a":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"amd","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"merchant-accelerators":{"alternatives":{"level":3,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/2488/000000248826000018/amd-20251227.htm","date":"2026-02-04","support":"AMD says it competes primarily against Intel and Nvidia in its data center products (Competition). Full date shown on the source."},{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","date":"2026-02-25","support":"NVIDIA lists AMD, Huawei and Intel as GPU and accelerator suppliers (Item 1, Competition); NVIDIA is a full alternative. Full date shown on the source."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://pureai.com/articles/2026/02/18/nvidia-expands-multiyear-ai-infrastructure-deal-with-meta.aspx","date":"2026-02-18","support":"Meta, an AMD Instinct customer, is also deploying millions of NVIDIA Blackwell and Rubin GPUs: the fallback is already qualified. Full date shown on the source."},{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","date":"2026-02-25","support":"NVIDIA has had supply lead times of more than 12 months (Item 1A), the main limit on a switch within a year. Full date shown on the source."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","date":"2026-02-25","support":"NVIDIA's Data Center Compute revenue was $162,361 million in fiscal 2026, many times AMD's accelerator volume. Full date shown on the source."},{"url":"https://www.sec.gov/Archives/edgar/data/2488/000000248826000018/amd-20251227.htm","date":"2026-02-04","support":"AMD's whole Data Center segment was $16,635 million in 2025; NVIDIA exceeds it, limited only by supply. Full date shown on the source."}]}},"host-cpus":{"alternatives":{"level":2,"evidence":[{"url":"https://www.theregister.com/2026/02/13/amd_intel_market_share/","date":"2026-02-13","support":"Intel held just over 71% of server processors in Q4 2025 (Mercury Research). Full date shown on the source."},{"url":"https://newsroom.arm.com/blog/half-of-compute-shipped-to-top-hyperscalers-in-2025-will-be-arm-based","date":"2025-04-01","support":"Arm says close to 50% of compute shipped to top hyperscalers in 2025 is Arm-based (Graviton, Axion, Cobalt, Grace). Full date shown on the source."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.khaleejtimes.com/business/tech/intel-amd-sign-long-term-server-cpu-deals-with-chinese-clients-as-prices-surge-sources-say","date":"2026-07-23","support":"Reuters: Intel lead times reached as long as six months for some server CPUs in 2026. Full date shown on the source."},{"url":"https://www.intel.com/content/www/us/en/newsroom/news/artificial-intelligence/new-intel-xeon-6-cpus-maximize-gpu-ai-performance.html","date":"2025-05-22","support":"Intel says Xeon 6 is the host CPU of NVIDIA DGX B300, so AI-server platforms already support Xeon hosts. Full date shown on the source."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.theregister.com/2026/02/13/amd_intel_market_share/","date":"2026-02-13","support":"Intel ships most server processors (just over 71% in Q4 2025), ample capability to absorb AMD's AI-host share. Full date shown on the source."},{"url":"https://www.khaleejtimes.com/business/tech/intel-amd-sign-long-term-server-cpu-deals-with-chinese-clients-as-prices-surge-sources-say","date":"2026-07-23","support":"Intel's CEO said demand 'continues to run ahead of supply' (April 2026): a small capacity gap. Full date shown on the source."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"amd","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"merchant-accelerators":{"alternatives":{"level":3,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/2488/000000248826000018/amd-20251227.htm","date":"2026-02-04","support":"AMD's 10-K says it competes primarily against Intel and Nvidia in data centre, plus smaller fabless accelerator makers (Competition). Date form: full filing date from the EDGAR index."},{"url":"https://www.anthropic.com/news/expanding-our-use-of-google-cloud-tpus-and-services","date":"2025-10-23","support":"Anthropic runs on Google TPUs, Amazon Trainium and NVIDIA GPUs, so TPU and Trainium capacity is also qualified for large AI workloads. Date form: full date shown."},{"url":"https://investor.nvidia.com/news/press-release-details/2026/Meta-Builds-AI-Infrastructure-With-NVIDIA/","date":"2026-02-17","support":"NVIDIA's release covers 'millions' of Blackwell and Rubin GPUs for Meta, announced a week before Meta's AMD deal, so NVIDIA is already qualified. Date form: full date shown (NVIDIA release)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://investor.nvidia.com/news/press-release-details/2026/Meta-Builds-AI-Infrastructure-With-NVIDIA/","date":"2026-02-17","support":"Meta signed NVIDIA (17 February) and AMD (24 February) multi-year deals within a week, so the NVIDIA path needs no new qualification at Meta. Date form: full date shown (NVIDIA release)."},{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","date":"2026-02-25","support":"NVIDIA says it has seen supply lead times over 12 months in shortages (Item 1A), which bounds how fast replacement volume can arrive. Date form: full filing date from the EDGAR index."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","date":"2026-02-25","support":"NVIDIA's Compute & Networking segment was $193,479 million in fiscal 2026, large enough to absorb AMD's volume subject to packaging supply. Date form: full filing date from the EDGAR index."},{"url":"https://the-decoder.com/nvidia-sets-new-mlperf-records-with-288-gpus-while-amd-and-intel-focus-on-different-battles/","date":"2026-04-02","support":"MI355X matched NVIDIA's B200 in some single-node tests but trailed the B300 on GPT-OSS-120B (MLPerf Inference v6.0), so the alternative is at least as capable. Date form: full date shown."}]}},"host-cpus":{"alternatives":{"level":2,"evidence":[{"url":"https://hothardware.com/news/amd-record-x86-market-share-cut-intels-chip-lead","date":"2026-08-21","support":"Intel holds 65.5% of x86 server CPU share against AMD's 34.5% (Mercury Research, Q2 2026), so Intel is a qualified alternative at scale. Date form: full date shown (reports Mercury Research data)."},{"url":"https://www.sec.gov/Archives/edgar/data/0001973239/000197323926000062/exhibit992fye26q431-marx26.htm","date":"2026-05-06","support":"Arm says it now has about 50% share of CPU compute among top hyperscalers, so Arm-based hosts are a further alternative. Date form: full filing date from the EDGAR index; the document itself shows no date."},{"url":"https://investor.nvidia.com/news/press-release-details/2026/Meta-Builds-AI-Infrastructure-With-NVIDIA/","date":"2026-02-17","support":"NVIDIA calls Meta's rollout the first large-scale Grace-only deployment, with potential large-scale Vera deployment in 2027. Date form: full date shown (NVIDIA release)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://hothardware.com/news/amd-record-x86-market-share-cut-intels-chip-lead","date":"2026-08-21","support":"AMD's server unit share rose 7.3 points in a year, which shows buyers can move volume between Intel and AMD within about a year. Date form: full date shown (reports Mercury Research data)."},{"url":"https://www.intc.com/news-events/press-releases/detail/1776/","date":"2026-07-23","support":"Intel says it is raising spending on equipment, clean room space and substrates to add capacity (Q2 2026 release). Date form: full date shown."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/50863/000005086326000011/intc-20251227.htm","date":"2026-01-23","support":"Intel says server revenue was limited by supply in Q4 2025, mainly Intel 7 and Intel 3 wafers, with constraints persisting into 2026. Date form: full filing date from the EDGAR index."},{"url":"https://www.intc.com/news-events/press-releases/detail/1776/","date":"2026-07-23","support":"Intel lists industry-wide substrate and memory shortages as a risk to meeting demand (Q2 2026 release): the best alternative cannot easily absorb more. Date form: full date shown."}]}}}}},"alphabet":{"a":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"alphabet","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"custom-accelerators":{"alternatives":{"level":3,"evidence":[{"url":"https://www.anthropic.com/news/expanding-our-use-of-google-cloud-tpus-and-services","date":"2025-10-23","support":"Anthropic, the largest outside TPU user, says it uses three chip platforms: TPUs, Trainium and NVIDIA GPUs. Full date shown on the source."},{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","date":"2026-02-25","support":"NVIDIA lists Alphabet with Amazon, Microsoft and others as cloud companies designing their own accelerators (Item 1). Full date shown on the source."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://www.anthropic.com/news/expanding-our-use-of-google-cloud-tpus-and-services","date":"2025-10-23","support":"Anthropic's TPU expansion brings well over a gigawatt of capacity online in 2026; replacing it takes over a year. Full date shown on the source."},{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","date":"2026-02-25","support":"NVIDIA has had supply lead times of more than 12 months (Item 1A), so GPU replacement capacity takes 1-2 years. Full date shown on the source."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://www.anthropic.com/news/expanding-our-use-of-google-cloud-tpus-and-services","date":"2025-10-23","support":"Anthropic alone plans up to one million TPUs, well over a gigawatt online in 2026. Full date shown on the source."},{"url":"https://www.aboutamazon.com/news/aws/aws-project-rainier-ai-trainium-chips-compute-cluster","date":"undated","support":"The largest Trainium build delivered nearly half a million Trainium2 chips; GPUs match TPUs but supply is committed: material gap. The page shows no publication date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"alphabet","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"custom-accelerators":{"alternatives":{"level":3,"evidence":[{"url":"https://www.anthropic.com/news/expanding-our-use-of-google-cloud-tpus-and-services","date":"2025-10-23","support":"Anthropic runs Claude across Google TPUs, Amazon Trainium and NVIDIA GPUs, so those platforms are the qualified alternatives to TPUs. Date form: full date shown."},{"url":"https://ir.amd.com/news-events/press-releases/detail/1279/","date":"2026-02-24","support":"AMD and Meta agree up to 6 GW of Instinct GPUs, the first gigawatt shipping from H2 2026: AMD is a further large-scale option. Date form: full date shown."},{"url":"https://www.theregister.com/systems/2026/05/22/alibaba-just-admitted-its-struggling-to-keep-up-with-rival-chipmakers-and-ai-shops/5244665","date":"2026-05-22","support":"The Register: Nvidia says AWS alone will rack one million of its GPUs this year, so NVIDIA supply reaches hyperscaler scale. Date form: full date shown."}]},"switchingTime":{"level":4,"evidence":[{"url":"https://www.sdxcentral.com/news/broadcom-secures-deal-with-google-for-next-gen-ai-racks-and-custom-tpus/","date":"2026-04-07","support":"Broadcom is to develop and supply Google's custom TPU silicon, with a supply assurance agreement for Google's next-generation racks through 2031. Date form: full date shown."},{"url":"https://www.anthropic.com/news/expanding-our-use-of-google-cloud-tpus-and-services","date":"2025-10-23","support":"Anthropic's October 2025 TPU expansion brings 'well over a gigawatt' online only during 2026, about a year of lead time for one customer's share. Date form: full date shown."},{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","date":"2026-02-25","support":"NVIDIA says it has long-term supply and capacity commitments and has seen lead times over 12 months, so shifting TPU volume to GPUs is slow. Date form: full filing date from the EDGAR index."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://counterpointresearch.com/de/insights/AI-Server-Compute-ASIC-Shipments-to-Triple-by-2027","date":"2026-01","support":"Counterpoint: Google held 64% of the AI server compute ASIC market in 2024, falling to 52% by 2027 but remaining the volume base. Date form: month only, because the page shows 26 January but its dateline year is inconsistent."},{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","date":"2026-02-25","support":"NVIDIA says it relies on CoWoS packaging and has locked in long-term supply and capacity agreements, which limits how fast it can add TPU-sized volume. Date form: full filing date from the EDGAR index."}]}}}}},"amazon":{"a":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"amazon","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"custom-accelerators":{"alternatives":{"level":3,"evidence":[{"url":"https://www.anthropic.com/news/expanding-our-use-of-google-cloud-tpus-and-services","date":"2025-10-23","support":"Anthropic, Trainium's main external user, also uses Google TPUs and NVIDIA GPUs (company says). Full date shown on the source."},{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","date":"2026-02-25","support":"NVIDIA lists Amazon among cloud companies designing their own accelerators (Item 1, Competition). Full date shown on the source."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://www.aboutamazon.com/news/aws/aws-project-rainier-ai-trainium-chips-compute-cluster","date":"undated","support":"AWS says Project Rainier became fully operational less than one year after announcement; rebuilding it elsewhere takes similar time. The page shows no publication date."},{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","date":"2026-02-25","support":"NVIDIA has had supply lead times of more than 12 months (Item 1A). Full date shown on the source."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://www.aboutamazon.com/news/aws/aws-project-rainier-ai-trainium-chips-compute-cluster","date":"undated","support":"Project Rainier delivered nearly half a million Trainium2 chips, with Claude expected on more than one million. The page shows no publication date."},{"url":"https://www.anthropic.com/news/expanding-our-use-of-google-cloud-tpus-and-services","date":"2025-10-23","support":"Google TPUs (up to one million for Anthropic) could cover it in capability, but that capacity is already committed. Full date shown on the source."}]}},"host-cpus":{"alternatives":{"level":2,"evidence":[{"url":"https://www.intel.com/content/www/us/en/newsroom/news/artificial-intelligence/new-intel-xeon-6-cpus-maximize-gpu-ai-performance.html","date":"2025-05-22","support":"Intel says a Xeon 6 is host CPU in NVIDIA DGX B300, showing x86 hosts qualified for AI servers. Full date shown on the source."},{"url":"https://www.theregister.com/2026/02/13/amd_intel_market_share/","date":"2026-02-13","support":"Intel and AMD supplied about 71% and 28.8% of server processors in Q4 2025 (Mercury Research). Full date shown on the source."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.khaleejtimes.com/business/tech/intel-amd-sign-long-term-server-cpu-deals-with-chinese-clients-as-prices-surge-sources-say","date":"2026-07-23","support":"Reuters: Intel lead times reached six months for some server CPUs in 2026. Full date shown on the source."},{"url":"https://www.intel.com/content/www/us/en/newsroom/news/artificial-intelligence/new-intel-xeon-6-cpus-maximize-gpu-ai-performance.html","date":"2025-05-22","support":"Xeon 6 hosts are already used in AI servers, so a switch is a platform requalification of months. Full date shown on the source."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.theregister.com/2026/02/13/amd_intel_market_share/","date":"2026-02-13","support":"x86 suppliers ship most server CPUs; Graviton's small AI-host volume is within their capability. Full date shown on the source."},{"url":"https://www.khaleejtimes.com/business/tech/intel-amd-sign-long-term-server-cpu-deals-with-chinese-clients-as-prices-surge-sources-say","date":"2026-07-23","support":"Intel's CEO said demand 'continues to run ahead of supply', a small capacity gap in 2026. Full date shown on the source."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"amazon","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"custom-accelerators":{"alternatives":{"level":3,"evidence":[{"url":"https://www.anthropic.com/news/expanding-our-use-of-google-cloud-tpus-and-services","date":"2025-10-23","support":"Anthropic uses Google TPUs, Amazon Trainium and NVIDIA GPUs, so TPUs and GPUs are qualified for the same workloads. Date form: full date shown."},{"url":"https://www.theregister.com/systems/2026/05/22/alibaba-just-admitted-its-struggling-to-keep-up-with-rival-chipmakers-and-ai-shops/5244665","date":"2026-05-22","support":"The Register: Nvidia says AWS alone will rack one million of its GPUs this year, so AWS already runs NVIDIA at scale. Date form: full date shown."},{"url":"https://www.theregister.com/2026/04/29/amazon_chips_20b_business/","date":"2026-04-30","support":"Trainium2 'has largely sold out' and Trainium3 is nearly fully subscribed, per Amazon's CEO, so demand already exceeds Trainium supply. Date form: full date shown on the page (the URL carries 29 April)."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://www.theregister.com/2026/04/29/amazon_chips_20b_business/","date":"2026-04-30","support":"Trainium4 is about 18 months from broad availability with much already reserved; OpenAI's ~2 GW ramps in 2027 and Anthropic has committed to up to 5 GW. Date form: full date shown on the page (the URL carries 29 April)."},{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","date":"2026-02-25","support":"NVIDIA says it has seen supply lead times over 12 months (Item 1A), so GPU replacement capacity would arrive over one to two years. Date form: full filing date from the EDGAR index."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.theregister.com/2026/04/29/amazon_chips_20b_business/","date":"2026-04-30","support":"Amazon's CEO claims Trainium2 has about 30% better price-performance than comparable GPUs: a company claim of a cost edge, not a capability gap. Date form: full date shown on the page (the URL carries 29 April)."},{"url":"https://www.theregister.com/systems/2026/05/22/alibaba-just-admitted-its-struggling-to-keep-up-with-rival-chipmakers-and-ai-shops/5244665","date":"2026-05-22","support":"Nvidia says AWS will rack one million of its GPUs this year, so the alternative has the scale to serve AWS. Date form: full date shown."}]}},"host-cpus":{"alternatives":{"level":2,"evidence":[{"url":"https://hothardware.com/news/amd-record-x86-market-share-cut-intels-chip-lead","date":"2026-08-21","support":"Mercury Research: AMD 34.5% and Intel 65.5% of x86 server CPU share in Q2 2026, so two x86 suppliers can serve AWS. Date form: full date shown (reports Mercury Research data)."},{"url":"https://newsroom.arm.com/blog/half-of-compute-shipped-to-top-hyperscalers-in-2025-will-be-arm-based","date":"2025-04-01","support":"Arm names Graviton, Axion, Cobalt and Grace as Neoverse-based; other Arm hosts exist but are captive to their makers. Date form: full date shown."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.theregister.com/2026/04/29/amazon_chips_20b_business/","date":"2026-04-30","support":"Amazon's CEO says 98% of the top 1,000 EC2 customers use Graviton and compares it with 'other x86 processors', so x86 is the existing fallback. Date form: full date shown on the page (the URL carries 29 April)."},{"url":"https://www.intc.com/news-events/press-releases/detail/1776/","date":"2026-07-23","support":"Intel reports server demand up and capacity being added (Q2 2026 release), though supply stays tight. Date form: full date shown."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.theregister.com/2026/04/29/amazon_chips_20b_business/","date":"2026-04-30","support":"Amazon's CEO claims Graviton beats other x86 processors on price-performance by up to 40%: a company claim of a small edge. Date form: full date shown on the page (the URL carries 29 April)."},{"url":"https://hothardware.com/news/amd-record-x86-market-share-cut-intels-chip-lead","date":"2026-08-21","support":"AMD has gained 7.3 points of x86 server unit share in a year, so x86 supply is expanding. Date form: full date shown (reports Mercury Research data)."}]}}}}},"marvell":{"a":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"marvell","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"custom-accelerators":{"alternatives":{"level":2,"evidence":[{"url":"https://www.taipeitimes.com/News/biz/archives/2025/05/30/2003837724","date":"2025-05-30","support":"Alchip is reportedly designing AWS's 3nm Trainium 3, Marvell's main custom programme customer. Full date shown on the source."},{"url":"https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm","date":"2026-03-11","support":"Marvell names Alchip and GUC among ASIC design competitors (Item 1, Competition), with Broadcom also in the field. Full date shown on the source."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.taipeitimes.com/News/biz/archives/2025/05/30/2003837724","date":"2025-05-30","support":"Alchip expects Trainium 3 small-volume production at end-2025 and significant ramp in 2026, within a year. Full date shown on the source."},{"url":"https://www.techpartner.news/news/google-preparing-to-partner-with-mediatek-on-next-ai-chip-615763","date":"2025-03-18","support":"Owners can add design partners: Google is preparing MediaTek alongside Broadcom (Reuters). Full date shown on the source."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1730168/000173016825000116/avgo-11022025x8kxex99.htm","date":"2025-12-11","support":"Broadcom guided Q1 fiscal 2026 AI revenue to $8.2 billion from custom accelerators and switches. Full date shown on the source."},{"url":"https://www.taipeitimes.com/News/biz/archives/2025/05/30/2003837724","date":"2025-05-30","support":"Alchip helps customers secure TSMC CoWoS capacity for 3nm AI chips: a small gap. Full date shown on the source."}]}},"networking-silicon":{"alternatives":{"level":3,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm","date":"2026-03-11","support":"Marvell names Broadcom and Cisco in switching and others in electro-optics as competitors (Item 1, Competition). Full date shown on the source."},{"url":"https://www.sec.gov/Archives/edgar/data/1807794/000162828026043303/crdo-20260502.htm","date":"2026-06-15","support":"Credo lists Broadcom, Marvell and Astera Labs as principal competitors (Item 1, Competition). Full date shown on the source."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1807794/000162828026043303/crdo-20260502.htm","date":"2026-06-15","support":"Customer evaluation of such interconnect chips may take several months or more than a year (Item 1A). Full date shown on the source."},{"url":"https://investor.marvell.com/news-events/press-releases/detail/121/marvell-expands-strategic-collaboration-with-aws-to-enable-accelerated-infrastructure-for-ai-in-the-cloud","date":"2024-12-02","support":"Marvell's AWS deal covers optical and AEC DSPs, PCIe retimers and switching, each also sourced from rivals. Full date shown on the source."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.sdxcentral.com/news/broadcom-debuts-taurus-dsp-to-power-16t-optical-modules-for-ai-networks/","date":"2026-03-11","support":"Broadcom is sampling a 400G-per-lane DSP for 1.6T modules, so rivals are at the frontier. Full date shown on the source."},{"url":"https://investor.marvell.com/news-events/press-releases/detail/1013/marvell-ushers-in-the-1-6t-era-with-expanded-optical-dsp-platform-portfolio-redefining-ai-data-center-end-to-end-connectivity","date":"2026-03-12","support":"Marvell says its Ara 1.6T DSP ships in mass volume; evidence on rivals' volume is thin. Full date shown on the source."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"marvell","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"custom-accelerators":{"alternatives":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm","date":"2026-03-11","support":"Marvell's own 10-K lists Alchip, Global Unichip, MediaTek, Socionext and Broadcom among direct competitors (Competition). Date form: full filing date from the EDGAR index."},{"url":"https://counterpointresearch.com/de/insights/AI-Server-Compute-ASIC-Shipments-to-Triple-by-2027","date":"2026-01","support":"Counterpoint: Alchip is expected to regain position via AWS; MediaTek has a TPU v8x partnership; Broadcom leads design-partner share. Date form: month only, because the page shows 26 January but its dateline year is inconsistent."},{"url":"https://www.sdxcentral.com/news/broadcom-secures-deal-with-google-for-next-gen-ai-racks-and-custom-tpus/","date":"2026-04-07","support":"MediaTek already has orders for v7e versions of Google's Ironwood TPU and will support packaging for TPU v8x, so another partner is in production. Date form: full date shown."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://siliconanalysts.com/analysis/custom-ai-chip-economics-tapeout-speed-vs-cost-2026","date":"2026-08-27","support":"A new design needs 18 to 36 months to first silicon plus 6 to 12 months of bring-up. Date form: full date shown."},{"url":"https://futurumgroup.com/insights/jalapeo-in-nine-months-did-ai-just-break-chip-design-timelines/","date":"2026-06-26","support":"Concept to tape-out in 18 to 24 months 'is considered fast' for a frontier accelerator. Date form: full date shown."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://counterpointresearch.com/de/insights/AI-Server-Compute-ASIC-Shipments-to-Triple-by-2027","date":"2026-01","support":"Broadcom is projected at about 60% of design-partner share in 2027, so the largest alternative has ample scale. Date form: month only, because the page shows 26 January but its dateline year is inconsistent."},{"url":"https://www.alchip.com/en/Newsroom/Alchip_2025_Q4_financial_results","date":"2026-03-18","support":"Alchip says its 3nm production programmes increase from Q2 2026; HPC and AI were 83% of 2025 revenue (company release). Date form: full date shown."}]}},"networking-silicon":{"alternatives":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm","date":"2026-03-11","support":"Marvell's 10-K names Broadcom, Credo, Astera Labs, Semtech, MACOM and Cisco among direct competitors (Competition). Date form: full filing date from the EDGAR index."},{"url":"https://semifundamental.substack.com/p/active-electrical-cable-aec-fundamentals","date":"2026-02-05","support":"Independent analysis: Marvell's retimers are the most widely adopted by AEC integrators, ahead of Broadcom's, with Astera and Credo also selling. Date form: full date shown (independent newsletter analysis)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://semifundamental.substack.com/p/active-electrical-cable-aec-fundamentals","date":"2026-02-05","support":"Qualifying a new retimer with a major cloud customer typically takes about 6 to 9 months of unit tests, system tests and pilots. Date form: full date shown (independent newsletter analysis)."},{"url":"https://www.theregister.com/networks/2026/06/02/marvell-enters-the-ai-network-fray-with-1024-tbps-switch-silicon/5250180","date":"2026-06-02","support":"Broadcom's Tomahawk 6 has shipped since 2025 and Cisco's G300 is shipping, so a second large source is already installed. Date form: full date shown."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://semifundamental.substack.com/p/active-electrical-cable-aec-fundamentals","date":"2026-02-05","support":"Marvell is described as only slightly ahead of Credo on 1.6T retimer maturity and priced 15-20% lower than Credo, so alternatives are close. Date form: full date shown (independent newsletter analysis)."},{"url":"https://www.theregister.com/networks/2026/06/02/marvell-enters-the-ai-network-fray-with-1024-tbps-switch-silicon/5250180","date":"2026-06-02","support":"The Register: Broadcom and Cisco already ship 102.4T switch silicon, so Marvell is not ahead of its alternatives. Date form: full date shown."}]}}}}},"intel":{"a":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"intel","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"custom-accelerators":{"alternatives":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/50863/000005086326000011/intc-20251227.htm","date":"2026-01-23","support":"Intel says 'we also compete with Broadcom in the custom ASICs development market' (Item 1, DCAI). Full date shown on the source."},{"url":"https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm","date":"2026-03-11","support":"Marvell names Alchip and GUC among ASIC design competitors (Item 1, Competition): several qualified design partners. Full date shown on the source."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://www.taipeitimes.com/News/biz/archives/2025/05/30/2003837724","date":"2025-05-30","support":"A design-partner change means a new chip cycle; Alchip's AWS 3nm chip moves from small-volume production to ramp over about a year (analogy). Full date shown on the source."},{"url":"https://www.sec.gov/Archives/edgar/data/1807794/000162828026043303/crdo-20260502.htm","date":"2026-06-15","support":"Chip design cycles from engagement to volume typically run two to three years (Credo, Item 1A; analogy). Full date shown on the source."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://d1io3yog0oux5.cloudfront.net/_3ccf9cf30376bf77bbb27f582e51d00d/intel/db/887/9159/prepared_remarks/Intel-4Q2025-Earnings-Call+1+(1).pdf","date":"2026-01-22","support":"The company says its custom ASIC run rate passed $1 billion in Q4 2025, driven by AI networking demand (CFO remarks, p. 5). Full date shown on the source."},{"url":"https://www.sec.gov/Archives/edgar/data/1730168/000173016825000116/avgo-11022025x8kxex99.htm","date":"2025-12-11","support":"Broadcom guided Q1 fiscal 2026 AI revenue to $8.2 billion from custom accelerators and switches: alternatives are far larger, a small gap. Full date shown on the source."}]}},"host-cpus":{"alternatives":{"level":2,"evidence":[{"url":"https://www.theregister.com/2026/02/13/amd_intel_market_share/","date":"2026-02-13","support":"AMD reached 28.8% of server processors in Q4 2025, up 3.1 points (Mercury Research). Full date shown on the source."},{"url":"https://newsroom.arm.com/blog/half-of-compute-shipped-to-top-hyperscalers-in-2025-will-be-arm-based","date":"2025-04-01","support":"Arm says close to 50% of compute shipped to top hyperscalers in 2025 is Arm-based, including NVIDIA Grace. Full date shown on the source."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.khaleejtimes.com/business/tech/intel-amd-sign-long-term-server-cpu-deals-with-chinese-clients-as-prices-surge-sources-say","date":"2026-07-23","support":"Reuters: server CPU lead times reached six months for some products and supply deals run about a year. Full date shown on the source."},{"url":"https://www.intel.com/content/www/us/en/newsroom/news/artificial-intelligence/new-intel-xeon-6-cpus-maximize-gpu-ai-performance.html","date":"2025-05-22","support":"Intel says Xeon 6 hosts NVIDIA DGX B300; replacing it means requalifying a host platform, typically months. Full date shown on the source."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://www.theregister.com/2026/02/13/amd_intel_market_share/","date":"2026-02-13","support":"With Intel just over 71% and AMD 28.8% of server units, AMD would need a large capacity increase. Full date shown on the source."},{"url":"https://www.khaleejtimes.com/business/tech/intel-amd-sign-long-term-server-cpu-deals-with-chinese-clients-as-prices-surge-sources-say","date":"2026-07-23","support":"Server CPU demand ran ahead of supply in 2026; some prices in China rose more than 40% since January. Full date shown on the source."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"intel","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"host-cpus":{"alternatives":{"level":2,"evidence":[{"url":"https://hothardware.com/news/amd-record-x86-market-share-cut-intels-chip-lead","date":"2026-08-21","support":"AMD holds 34.5% of x86 server CPU share in Q2 2026 (Mercury Research), so AMD is qualified in volume. Date form: full date shown (reports Mercury Research data)."},{"url":"https://www.sec.gov/Archives/edgar/data/0001973239/000197323926000062/exhibit992fye26q431-marx26.htm","date":"2026-05-06","support":"Arm says about 50% of CPU compute among top hyperscalers is now on Arm, so Arm-based hosts are further alternatives. Date form: full filing date from the EDGAR index; the document itself shows no date."},{"url":"https://investor.nvidia.com/news/press-release-details/2026/Meta-Builds-AI-Infrastructure-With-NVIDIA/","date":"2026-02-17","support":"NVIDIA calls Meta's rollout the first large-scale Grace-only deployment, with potential large-scale Vera deployment in 2027. Date form: full date shown (NVIDIA release)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://ir.amd.com/news-events/press-releases/detail/1295/amd-reports-second-quarter-2026-financial-results","date":"2026-08-04","support":"AMD says EPYC demand is accelerating and expects Data Center sales to grow faster in the second half of 2026 (Q2 2026 results). Date form: full date shown."},{"url":"https://www.sec.gov/Archives/edgar/data/50863/000005086326000011/intc-20251227.htm","date":"2026-01-23","support":"Intel's own supply limits already hold back its server revenue (Q4 2025 10-K), so buyers already face constrained Intel supply. Date form: full filing date from the EDGAR index."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/2488/000000248826000018/amd-20251227.htm","date":"2026-02-04","support":"AMD says TSMC makes all its wafers at 7 nm or smaller and a TSMC shortfall could have a material adverse effect (Item 1A). Date form: full filing date from the EDGAR index."},{"url":"https://ir.amd.com/news-events/press-releases/detail/1295/amd-reports-second-quarter-2026-financial-results","date":"2026-08-04","support":"AMD's Data Center revenue was $6,718 million in Q2 2026, up 107%, so its capacity is growing fast. Date form: full date shown."}]}},"custom-accelerators":{"alternatives":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/50863/000005086326000011/intc-20251227.htm","date":"2026-01-23","support":"Intel's 10-K says it competes with Broadcom in the custom ASICs development market and names hyperscalers that design their own silicon (Item 1, DCAI). Date form: full filing date from the EDGAR index."},{"url":"https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm","date":"2026-03-11","support":"Marvell's 10-K lists Alchip, Global Unichip, MediaTek, Socionext and Broadcom among direct competitors, so several design partners exist. Date form: full filing date from the EDGAR index."},{"url":"https://counterpointresearch.com/de/insights/AI-Server-Compute-ASIC-Shipments-to-Triple-by-2027","date":"2026-01","support":"Counterpoint names Broadcom (about 60% share in 2027), Marvell (about 8%), Alchip and MediaTek as the other design partners for hyperscalers' in-house chips. Date form: month only, because the page shows 26 January but its dateline year is inconsistent."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://siliconanalysts.com/analysis/custom-ai-chip-economics-tapeout-speed-vs-cost-2026","date":"2026-08-27","support":"A new custom design takes 18 to 36 months from architecture freeze to first silicon, plus 6 to 12 months of bring-up and validation. Date form: full date shown."},{"url":"https://d1io3yog0oux5.cloudfront.net/_3ccf9cf30376bf77bbb27f582e51d00d/intel/db/887/9159/prepared_remarks/Intel-4Q2025-Earnings-Call+1+(1).pdf","date":"2026-01-22","support":"Intel offers design services, IP building blocks and manufacturing to customers seeking purpose-built silicon, so each win is a custom design programme (inference). Date form: full date shown in the document (22 January 2026)."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://d1io3yog0oux5.cloudfront.net/_3ccf9cf30376bf77bbb27f582e51d00d/intel/db/887/9159/prepared_remarks/Intel-4Q2025-Earnings-Call+1+(1).pdf","date":"2026-01-22","support":"Intel says its custom ASIC business reached an annualised revenue run rate 'greater than $1 billion' in Q4 2025, driven by AI networking demand. Date form: full date shown in the document (22 January 2026)."},{"url":"https://www.sec.gov/Archives/edgar/data/0001730168/000173016826000076/avgo-08022026x8kxex99.htm","date":"2026-09-02","support":"Broadcom reported AI semiconductor revenue of $16.7 billion in its fiscal Q3 2026 alone, so the largest alternative has far more capacity than Intel's ASIC base. Date form: full date from the release dateline."}]}}}}},"arm":{"a":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"arm","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"processor-ip":{"alternatives":{"level":5,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1973239/000197323926000097/arm-20260331.htm","date":"2026-05-26","support":"Arm names x86 and RISC-V as competing architectures (Item 3.D, competition risk factor); neither is at parity for licensed server cores (inference). Full date shown on the source."},{"url":"https://www.eetasia.com/risc-v-targets-data-centers-as-open-isa-pushes-beyond-embedded-systems/","date":"2026-07-30","support":"Server-class RISC-V systems are only 'starting to appear' in 2026 (Krste Asanovic, RISC-V International). Full date shown on the source."}]},"switchingTime":{"level":4,"evidence":[{"url":"https://www.eetasia.com/risc-v-targets-data-centers-as-open-isa-pushes-beyond-embedded-systems/","date":"2026-07-30","support":"Displacing incumbent ISAs in the data centre 'will take several generations of technology'. Full date shown on the source."},{"url":"https://www.intel.com/content/www/us/en/newsroom/news/artificial-intelligence/intel-and-nvidia-to-jointly-develop-ai-infrastructure-and-personal-computing-products.html","date":"2025-09-18","support":"Intel building NVIDIA-custom x86 CPUs, announced September 2025, is the one licensed-x86 route seen; others would buy x86 chips (inference). Full date shown on the source."}]},"capabilityGap":{"level":5,"evidence":[{"url":"https://newsroom.arm.com/blog/half-of-compute-shipped-to-top-hyperscalers-in-2025-will-be-arm-based","date":"2025-04-01","support":"Arm says close to 50% of compute shipped to top hyperscalers in 2025 is Arm-based (Graviton, Axion, Cobalt, Grace). Full date shown on the source."},{"url":"https://www.eetasia.com/risc-v-targets-data-centers-as-open-isa-pushes-beyond-embedded-systems/","date":"2026-07-30","support":"RISC-V server parts are at the first-product stage, so no alternative exists at scale. Full date shown on the source."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"arm","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"processor-ip":{"alternatives":{"level":4,"evidence":[{"url":"https://hothardware.com/news/sifive-pushes-risc-v-into-datacenters-bigsky-server-platform","date":"2026-08-24","support":"SiFive's 32-core RISC-V BigSky server is the first rackable 2U RISC-V server and ships in limited production quantities as a porting and validation platform. Date form: full date shown."},{"url":"https://www.sec.gov/Archives/edgar/data/0001973239/000197323926000062/exhibit992fye26q431-marx26.htm","date":"2026-05-06","support":"Arm says it has about 50% share of CPU compute among top hyperscalers, and more than half of its top 30 customers hold Total Access licences. Date form: full filing date from the EDGAR index; the document itself shows no date."}]},"switchingTime":{"level":4,"evidence":[{"url":"https://newsroom.arm.com/blog/half-of-compute-shipped-to-top-hyperscalers-in-2025-will-be-arm-based","date":"2025-04-01","support":"Arm says Graviton, Axion, Cobalt and NVIDIA Grace are all Neoverse-based, so moving them off Arm means new cores and a software port. Date form: full date shown."},{"url":"https://hothardware.com/news/sifive-pushes-risc-v-into-datacenters-bigsky-server-platform","date":"2026-08-24","support":"SiFive describes BigSky as built to help customers with software porting and system validation, so RISC-V server software is still being ported. Date form: full date shown."},{"url":"https://www.sec.gov/Archives/edgar/data/50863/000005086326000011/intc-20251227.htm","date":"2026-01-23","support":"The x86 fallback is supply-limited: Intel says server supply constraints persist into 2026. Date form: full filing date from the EDGAR index."}]},"capabilityGap":{"level":4,"evidence":[{"url":"https://hothardware.com/news/sifive-pushes-risc-v-into-datacenters-bigsky-server-platform","date":"2026-08-24","support":"The RISC-V server shown (32 P870-D cores at 2GHz) is a development-class system with 'demand outpacing supply'; no benchmarks are given. Date form: full date shown."},{"url":"https://www.sec.gov/Archives/edgar/data/0001973239/000197323926000062/exhibit992fye26q431-marx26.htm","date":"2026-05-06","support":"Arm reports its data-centre royalties more than doubled year over year and more than $2 billion of Arm AGI CPU demand across fiscal 2027-28. Date form: full filing date from the EDGAR index; the document itself shows no date."}]}}}}},"alibaba":{"a":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"alibaba","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"custom-accelerators":{"alternatives":{"level":3,"evidence":[{"url":"https://kfgo.com/?p=1285523","date":"2026-04-01","support":"IDC: Huawei shipped about 812,000 chips and Cambricon and Kunlunxin about 116,000 each in China in 2025. Full date shown on the source."},{"url":"https://www.thestandard.com.hk/innovation/article/311827/China-tells-tech-firms-to-stop-buying-Nvidias-AI-chips-FT-reports","date":"2025-09-17","support":"China's regulator told Alibaba and ByteDance to end testing and orders of Nvidia's RTX Pro 6000D (Reuters). Full date shown on the source."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://www.thestandard.com.hk/innovation/article/311827/China-tells-tech-firms-to-stop-buying-Nvidias-AI-chips-FT-reports","date":"2025-09-17","support":"With Nvidia purchases discouraged, Alibaba would have to port T-Head workloads to another domestic stack. Full date shown on the source."},{"url":"https://kfgo.com/?p=1285523","date":"2026-04-01","support":"Local projects reportedly carried implicit 'buy Chinese' directives, narrowing the fast options. Full date shown on the source."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://kfgo.com/?p=1285523","date":"2026-04-01","support":"T-Head's about 265,000 cards were second only to Huawei among Chinese vendors in 2025. Full date shown on the source."},{"url":"https://www.huawei.com/en/news/2025/9/hc-xu-keynote-speech","date":"2025-09-18","support":"Huawei says 'we don't have access to advanced process nodes', limiting how fast it can add volume. Full date shown on the source."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"alibaba","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"custom-accelerators":{"alternatives":{"level":3,"evidence":[{"url":"https://whbl.com/2026/04/01/chinese-chipmakers-claim-nearly-half-of-local-market-as-nvidias-lead-shrinks/","date":"2026-04-01","support":"IDC (via Reuters): T-Head shipped about 265,000 cards in China in 2025; Huawei about 812,000, Kunlunxin and Cambricon about 116,000 each, so other domestic suppliers are in volume. Date form: full date shown (Reuters report on IDC data, wire copy)."},{"url":"https://www.nst.com.my/amp/business/corporate/2026/07/1484039/china-plans-let-top-ai-firms-buy-limited-nvidia-h200-chips","date":"2026-07-09","support":"Beijing may allow about 200,000 or fewer NVIDIA H200 chips in total across Alibaba, ByteDance and DeepSeek, so NVIDIA is only a partial fallback. Date form: full date shown (Reuters story)."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://gulfnews.com/technology/huawei-to-double-output-of-top-ai-chip-as-nvidia-wavers-in-china-1.500287469","date":"2025-09-30","support":"Bloomberg: Huawei, the largest domestic supplier, plans about 600,000 910C chips in 2026, roughly double 2025, so spare domestic supply takes one to two years to appear. Date form: full date from the page's last-updated line (Bloomberg report)."},{"url":"https://www.theregister.com/systems/2026/05/22/alibaba-just-admitted-its-struggling-to-keep-up-with-rival-chipmakers-and-ai-shops/5244665","date":"2026-05-22","support":"The Register: Beijing has not let local buyers acquire Nvidia chips, and T-Head can't get much hardware made. Date form: full date shown."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://www.theregister.com/systems/2026/05/22/alibaba-just-admitted-its-struggling-to-keep-up-with-rival-chipmakers-and-ai-shops/5244665","date":"2026-05-22","support":"The Register: Chinese fabs can't yet match TSMC, so domestic capacity is the limit on any replacement. Date form: full date shown."},{"url":"https://gulfnews.com/technology/huawei-to-double-output-of-top-ai-chip-as-nvidia-wavers-in-china-1.500287469","date":"2025-09-30","support":"Bloomberg: SMIC is stuck at 7 nm and Huawei relies on stockpiled foreign HBM, so the best domestic alternative is itself supply-limited. Date form: full date from the page's last-updated line (Bloomberg report)."}]}}}}},"credo":{"a":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"credo","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"networking-silicon":{"alternatives":{"level":3,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1807794/000162828026043303/crdo-20260502.htm","date":"2026-06-15","support":"Credo names Broadcom, Marvell and Astera Labs as principal competitors (Item 1, Competition). Full date shown on the source."},{"url":"https://investor.marvell.com/news-events/press-releases/detail/121/marvell-expands-strategic-collaboration-with-aws-to-enable-accelerated-infrastructure-for-ai-in-the-cloud","date":"2024-12-02","support":"Marvell supplies AEC DSPs to AWS under a five-year agreement (company says). Full date shown on the source."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1807794/000162828026043303/crdo-20260502.htm","date":"2026-06-15","support":"Customer evaluation 'may take several months or more than a year' (Item 1A). Full date shown on the source."},{"url":"https://www.sec.gov/Archives/edgar/data/1736297/000173629726000010/alab-20251231.htm","date":"2026-02-20","support":"Astera says qualification 'may continue for several months or more' (Item 1A). Full date shown on the source."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://investor.marvell.com/news-events/press-releases/detail/121/marvell-expands-strategic-collaboration-with-aws-to-enable-accelerated-infrastructure-for-ai-in-the-cloud","date":"2024-12-02","support":"Marvell offers AEC DSPs, optical DSPs and retimers to a hyperscaler, covering Credo's main products. Full date shown on the source."},{"url":"https://www.sec.gov/Archives/edgar/data/1736297/000173629726000010/alab-20251231.htm","date":"2026-02-20","support":"Astera Labs lists Credo among its principal competitors, offering overlapping connectivity products. Full date shown on the source."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"credo","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"networking-silicon":{"alternatives":{"level":3,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/0001807794/000162828026043303/crdo-20260502.htm","date":"2026-06-15","support":"Credo's 10-K names Broadcom, Marvell and Astera Labs, plus various cable and optical transceiver suppliers, as principal competitors. Date form: full filing date from the EDGAR index."},{"url":"https://semifundamental.substack.com/p/active-electrical-cable-aec-fundamentals","date":"2026-02-05","support":"Independent analysis: Marvell retimers are most widely adopted by AEC integrators, Broadcom less so, Astera priced below Credo. Date form: full date shown (independent newsletter analysis)."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/0001807794/000162828026043303/crdo-20260502.htm","date":"2026-06-15","support":"Credo says design cycle from initial engagement to volume shipment is typically two to three years; customers may take several months to over a year to evaluate. Date form: full filing date from the EDGAR index."},{"url":"https://semifundamental.substack.com/p/active-electrical-cable-aec-fundamentals","date":"2026-02-05","support":"Qualifying with a major cloud customer typically takes 6 to 9 months of tests and pilots, before any volume ramp. Date form: full date shown (independent newsletter analysis)."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://semifundamental.substack.com/p/active-electrical-cable-aec-fundamentals","date":"2026-02-05","support":"Marvell is called slightly ahead of Credo on 1.6T retimer maturity and priced 15-20% lower; Credo is the incumbent with 'dominant shares previously'. Date form: full date shown (independent newsletter analysis)."},{"url":"https://www.sec.gov/Archives/edgar/data/0001807794/000162828026043303/crdo-20260502.htm","date":"2026-06-15","support":"Credo's top 10 customers were about 90% of fiscal 2026 revenue, so the volume to replace sits with a few buyers who have alternatives. Date form: full filing date from the EDGAR index."}]}}}}},"meta":{"a":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"meta","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"custom-accelerators":{"alternatives":{"level":3,"evidence":[{"url":"https://pureai.com/articles/2026/02/18/nvidia-expands-multiyear-ai-infrastructure-deal-with-meta.aspx","date":"2026-02-18","support":"NVIDIA will supply Meta with millions of Blackwell and Rubin GPUs under a multiyear deal. Full date shown on the source."},{"url":"https://newsroom.amd.com/news/amd-and-meta-announce-expanded-strategic-partnersh/","date":"2026-02-24","support":"AMD and Meta agreed to deploy up to 6 gigawatts of AMD Instinct GPUs (company says). Full date shown on the source."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://pureai.com/articles/2026/02/18/nvidia-expands-multiyear-ai-infrastructure-deal-with-meta.aspx","date":"2026-02-18","support":"Meta already runs NVIDIA GPUs at scale, so MTIA work can move to qualified GPUs within months. Full date shown on the source."},{"url":"https://newsroom.amd.com/news/amd-and-meta-announce-expanded-strategic-partnersh/","date":"2026-02-24","support":"AMD says shipments for Meta's first gigawatt begin in the second half of 2026. Full date shown on the source."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://ai.meta.com/blog/meta-training-inference-accelerator-AI-MTIA/","date":"2023-05-18","support":"Meta built MTIA because GPUs 'were not always optimal' for its recommendation workloads: an efficiency gap, not a capability one. Full date shown on the source."},{"url":"https://pureai.com/articles/2026/02/18/nvidia-expands-multiyear-ai-infrastructure-deal-with-meta.aspx","date":"2026-02-18","support":"Meta's NVIDIA deal covers training and inference across its data centres. Full date shown on the source."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"meta","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"custom-accelerators":{"alternatives":{"level":3,"evidence":[{"url":"https://investor.nvidia.com/news/press-release-details/2026/Meta-Builds-AI-Infrastructure-With-NVIDIA/","date":"2026-02-17","support":"NVIDIA's release covers 'millions' of Blackwell and Rubin GPUs for Meta under a multiyear partnership (17 February 2026). Date form: full date shown (NVIDIA release)."},{"url":"https://ir.amd.com/news-events/press-releases/detail/1279/","date":"2026-02-24","support":"AMD and Meta agree up to 6 GW of Instinct GPUs; the first gigawatt ships from H2 2026. Date form: full date shown."},{"url":"https://www.nst.com.my/amp/business/corporate/2025/11/1323188/meta-talks-spend-billions-googles-chips","date":"2025-11-26","support":"Reuters (citing The Information): Meta is in talks to spend billions on Google's TPUs from 2027 and rent them in 2026; a report, not a signed deal. Date form: full date shown (Reuters story citing The Information)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://investor.nvidia.com/news/press-release-details/2026/Meta-Builds-AI-Infrastructure-With-NVIDIA/","date":"2026-02-17","support":"Meta already buys NVIDIA GPUs by the million, so replacing MTIA volume needs no new supplier qualification. Date form: full date shown (NVIDIA release)."},{"url":"https://ir.amd.com/news-events/press-releases/detail/1279/","date":"2026-02-24","support":"Meta's AMD agreement is also in place: up to 6 GW, with the first gigawatt shipping from H2 2026. Date form: full date shown."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://investor.nvidia.com/news/press-release-details/2026/Meta-Builds-AI-Infrastructure-With-NVIDIA/","date":"2026-02-17","support":"NVIDIA's 'millions' of Blackwell and Rubin GPUs for Meta show the alternative is capable at Meta's scale. Date form: full date shown (NVIDIA release)."},{"url":"https://ir.amd.com/news-events/press-releases/detail/1279/","date":"2026-02-24","support":"AMD's first gigawatt for Meta is a custom MI450-based GPU with Helios racks, shipping from H2 2026. Date form: full date shown."}]}}}}},"bytedance":{"a":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"bytedance","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"custom-accelerators":{"alternatives":{"level":3,"evidence":[{"url":"https://www.thestandard.com.hk/innovation/article/311827/China-tells-tech-firms-to-stop-buying-Nvidias-AI-chips-FT-reports","date":"2025-09-17","support":"ByteDance had been testing Nvidia's RTX Pro 6000D before the regulator's order (Reuters), so it buys merchant GPUs. Full date shown on the source."},{"url":"https://kfgo.com/?p=1285523","date":"2026-04-01","support":"IDC: Huawei, T-Head, Cambricon and Kunlunxin all shipped domestic accelerators in China in 2025. Full date shown on the source."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://fortune.com/asia/2024/06/24/tiktok-owner-bytedance-talks-with-us-tech-firm-broadcom-about-tsmc-made-ai-chip","date":"2024-06-24","support":"Broadcom already supplies an older 7nm AI processor for ByteDance data centres (Bloomberg); volume is unconfirmed, evidence thin. Full date shown on the source."},{"url":"https://kfgo.com/?p=1285523","date":"2026-04-01","support":"Nvidia still shipped about 2.2 million cards in China in 2025, so qualified merchant supply exists. Full date shown on the source."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://kfgo.com/?p=1285523","date":"2026-04-01","support":"Nvidia's about 2.2 million and Huawei's about 812,000 cards in China dwarf any ByteDance in-house programme. Full date shown on the source."},{"url":"https://fortune.com/asia/2024/06/24/tiktok-owner-bytedance-talks-with-us-tech-firm-broadcom-about-tsmc-made-ai-chip","date":"2024-06-24","support":"ByteDance's custom chip is a Broadcom design; no source gives its 2025 volume (thin evidence). Full date shown on the source."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"bytedance","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"custom-accelerators":{"alternatives":{"level":3,"evidence":[{"url":"https://whbl.com/2026/04/01/chinese-chipmakers-claim-nearly-half-of-local-market-as-nvidias-lead-shrinks/","date":"2026-04-01","support":"IDC (via Reuters): Huawei, T-Head, Kunlunxin and Cambricon each shipped 116,000 to 812,000 cards in China in 2025, so several domestic suppliers are in volume. Date form: full date shown (Reuters report on IDC data, wire copy)."},{"url":"https://www.nst.com.my/amp/business/corporate/2026/07/1484039/china-plans-let-top-ai-firms-buy-limited-nvidia-h200-chips","date":"2026-07-09","support":"Beijing may allow about 200,000 or fewer NVIDIA H200 chips across Alibaba, ByteDance and DeepSeek, so NVIDIA is a partial fallback. Date form: full date shown (Reuters story)."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://technode.com/2026/02/11/bytedance-said-to-be-developing-ai-chip-in-talks-with-samsung-over-manufacturing/","date":"2026-02-11","support":"Reuters (via TechNode): ByteDance plans at least 100,000 units of an inference chip this year; ByteDance called the report inaccurate, so evidence on its programme is thin. Date form: full date shown (reports Reuters)."},{"url":"https://gulfnews.com/technology/huawei-to-double-output-of-top-ai-chip-as-nvidia-wavers-in-china-1.500287469","date":"2025-09-30","support":"Bloomberg: the largest domestic supplier plans about 600,000 910C chips in 2026, so spare domestic supply takes one to two years to appear. Date form: full date from the page's last-updated line (Bloomberg report)."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://www.theregister.com/systems/2026/05/22/alibaba-just-admitted-its-struggling-to-keep-up-with-rival-chipmakers-and-ai-shops/5244665","date":"2026-05-22","support":"The Register: Chinese fabs can't yet match TSMC and Beijing has not let buyers acquire Nvidia chips, so replacement capacity is limited. Date form: full date shown."},{"url":"https://www.nst.com.my/amp/business/corporate/2026/07/1484039/china-plans-let-top-ai-firms-buy-limited-nvidia-h200-chips","date":"2026-07-09","support":"Chinese officials had withheld approval of H200 purchases and were 'keen to nurture domestic suppliers' (Reuters, July 2026). Date form: full date shown (Reuters story)."}]}}}}},"astera-labs":{"a":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"astera-labs","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"networking-silicon":{"alternatives":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1736297/000173629726000010/alab-20251231.htm","date":"2026-02-20","support":"Principal competitors: Broadcom, Credo, Marvell, Microchip, Montage, Parade, Rambus (Item 1); not all match its PCIe 6 range (inference). Full date shown on the source."},{"url":"https://investor.marvell.com/news-events/press-releases/detail/139/marvell-expands-connectivity-portfolio-with-new-pcie-retimer-product-line-to-scale-compute-fabrics-of-accelerated-infrastructure","date":"2024-05-30","support":"Marvell launched PCIe Gen 6 Alaska P retimers, 16-lane sampling in May 2024. Full date shown on the source."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1736297/000173629726000010/alab-20251231.htm","date":"2026-02-20","support":"Customer qualification 'may continue for several months or more' (Item 1A). Full date shown on the source."},{"url":"https://investor.marvell.com/news-events/press-releases/detail/121/marvell-expands-strategic-collaboration-with-aws-to-enable-accelerated-infrastructure-for-ai-in-the-cloud","date":"2024-12-02","support":"Marvell already supplies PCIe retimers to AWS under a multi-year agreement (company says). Full date shown on the source."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://investor.marvell.com/news-events/press-releases/detail/139/marvell-expands-connectivity-portfolio-with-new-pcie-retimer-product-line-to-scale-compute-fabrics-of-accelerated-infrastructure","date":"2024-05-30","support":"Marvell's PCIe Gen 6 retimers match the generation of Astera's lead product: a small gap. Full date shown on the source."},{"url":"https://www.sec.gov/Archives/edgar/data/1736297/000173629726000010/alab-20251231.htm","date":"2026-02-20","support":"Seven named competitors offer overlapping products; evidence on their volumes is thin. Full date shown on the source."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"astera-labs","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"networking-silicon":{"alternatives":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1736297/000173629726000010/alab-20251231.htm","date":"2026-02-20","support":"Astera's 10-K names Broadcom, Credo, Marvell, Microchip, Montage, Parade and Rambus as principal competitors (Competition). Date form: full filing date from the EDGAR index."},{"url":"https://semifundamental.substack.com/p/active-electrical-cable-aec-fundamentals","date":"2026-02-05","support":"Independent analysis: Astera retimers are priced toward the low end of the AEC market and compete with Marvell and Broadcom parts. Date form: full date shown (independent newsletter analysis)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1736297/000173629726000010/alab-20251231.htm","date":"2026-02-20","support":"Astera says qualification 'may continue for several months or more' and a design win can take up to two years to reach volume shipments. Date form: full filing date from the EDGAR index."},{"url":"https://semifundamental.substack.com/p/active-electrical-cable-aec-fundamentals","date":"2026-02-05","support":"Qualifying with a major cloud customer typically takes about 6 to 9 months. Date form: full date shown (independent newsletter analysis)."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1736297/000173629726000010/alab-20251231.htm","date":"2026-02-20","support":"Astera's competitors, Broadcom, Marvell and Credo among them, overlap with some but not all of its products (Competition). Date form: full filing date from the EDGAR index."},{"url":"https://semifundamental.substack.com/p/active-electrical-cable-aec-fundamentals","date":"2026-02-05","support":"Astera's AEC retimers use legacy 12nm PCIe retimer technology and are priced below Credo to gain share, so rivals match its capability. Date form: full date shown (independent newsletter analysis)."}]}}}}},"hygon":{"a":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"hygon","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"host-cpus":{"alternatives":{"level":2,"evidence":[{"url":"https://www.khaleejtimes.com/business/tech/intel-amd-sign-long-term-server-cpu-deals-with-chinese-clients-as-prices-surge-sources-say","date":"2026-07-23","support":"Intel and AMD still supply server CPUs to Chinese customers under one-year volume commitments (Reuters). Full date shown on the source."},{"url":"https://www.huawei.com/en/news/2025/9/hc-xu-keynote-speech","date":"2025-09-18","support":"Huawei plans Kunpeng 950 server CPUs in 96- and 192-core models for Q1 2026 (keynote dateline 18 September 2025). Full date shown on the source."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.khaleejtimes.com/business/tech/intel-amd-sign-long-term-server-cpu-deals-with-chinese-clients-as-prices-surge-sources-say","date":"2026-07-23","support":"Reuters: lead times for some Intel server CPUs in China reached six months; deals cover about a year. Full date shown on the source."},{"url":"https://www.intel.com/content/www/us/en/newsroom/news/artificial-intelligence/new-intel-xeon-6-cpus-maximize-gpu-ai-performance.html","date":"2025-05-22","support":"Intel says a Xeon 6 hosts NVIDIA DGX B300: AI servers already take x86 hosts from other vendors. Full date shown on the source."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.khaleejtimes.com/business/tech/intel-amd-sign-long-term-server-cpu-deals-with-chinese-clients-as-prices-surge-sources-say","date":"2026-07-23","support":"Some server CPU prices in China rose more than 40% since January 2026: alternatives exist but supply is tight. Full date shown on the source."},{"url":"https://www.huawei.com/en/news/2025/9/hc-xu-keynote-speech","date":"2025-09-18","support":"Kunpeng's roadmap adds higher-core-count models, so capability is comparable; the gap is mainly supply. Full date shown on the source."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"hygon","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"host-cpus":{"alternatives":{"level":2,"evidence":[{"url":"https://sinolytics.de/global-business-news/blog/technology/china-cpu-self-sufficiency-policy-procurement/","date":"2026-07-06","support":"Sinolytics: Hygon and Huawei Kunpeng lead domestic CPUs, with Phytium, Zhaoxin and Loongson also supplying. Date form: full date shown."},{"url":"https://thenextweb.com/news/china-telecom-server-deal-huawei-kunpeng-1-7bn","date":"2026-06-24","support":"China Telecom's June 2026 tender gave 28,000 servers to the Kunpeng Arm platform and 12,000 to C86 servers, so alternatives are in volume. Date form: full date shown."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://thenextweb.com/news/china-telecom-server-deal-huawei-kunpeng-1-7bn","date":"2026-06-24","support":"Buyers already bought both Arm (Kunpeng) and x86-compatible (C86) servers in one tender, so a switch between them is a procurement choice. Date form: full date shown."},{"url":"https://heise.de/-10385441","date":"2025-05-15","support":"Hygon and AMD no longer formally work together and Hygon has been on the US Entity List since 2019, so Hygon is on its own roadmap. Date form: full date shown."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://heise.de/-10385441","date":"2025-05-15","support":"Hygon's 128-core C86-5G is estimated around Zen 3 level (roadmap-based), so Kunpeng and Intel/AMD parts are not far ahead. Date form: full date shown."},{"url":"https://sinolytics.de/global-business-news/blog/technology/china-cpu-self-sufficiency-policy-procurement/","date":"2026-07-06","support":"Sinolytics: domestic CPUs hold about 20% of China's overall CPU market, so most volume is still supplied by others. Date form: full date shown."}]}}}}},"baidu":{"a":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"baidu","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"custom-accelerators":{"alternatives":{"level":3,"evidence":[{"url":"https://kfgo.com/?p=1285523","date":"2026-04-01","support":"IDC: Huawei (about 812,000) and Cambricon (about 116,000) shipped comparable domestic accelerators in China in 2025. Full date shown on the source."},{"url":"https://www.thestandard.com.hk/innovation/article/311827/China-tells-tech-firms-to-stop-buying-Nvidias-AI-chips-FT-reports","date":"2025-09-17","support":"China's regulator told firms including ByteDance and Alibaba to end orders of Nvidia's RTX Pro 6000D (Reuters). Full date shown on the source."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://kfgo.com/?p=1285523","date":"2026-04-01","support":"Nvidia's most advanced products are cut off by U.S. controls, so Baidu would port to Ascend or Cambricon stacks. Full date shown on the source."},{"url":"https://www.thestandard.com.hk/innovation/article/311827/China-tells-tech-firms-to-stop-buying-Nvidias-AI-chips-FT-reports","date":"2025-09-17","support":"Chinese buyers stopped testing and ordering Nvidia's China-specific RTX Pro 6000D in September 2025. Full date shown on the source."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://kfgo.com/?p=1285523","date":"2026-04-01","support":"Kunlunxin's about 116,000 cards equal Cambricon's and are a seventh of Huawei's: alternatives have more volume. Full date shown on the source."},{"url":"https://www.huawei.com/en/news/2025/9/hc-xu-keynote-speech","date":"2025-09-18","support":"Huawei has deployed more than 300 Atlas 900 A3 SuperPoDs (keynote dateline 18 September 2025). Full date shown on the source."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"baidu","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"custom-accelerators":{"alternatives":{"level":3,"evidence":[{"url":"https://whbl.com/2026/04/01/chinese-chipmakers-claim-nearly-half-of-local-market-as-nvidias-lead-shrinks/","date":"2026-04-01","support":"IDC (via Reuters): Kunlunxin shipped about 116,000 cards in China in 2025; Huawei about 812,000, T-Head about 265,000 and Cambricon about 116,000, so other domestic suppliers are in volume. Date form: full date shown (Reuters report on IDC data, wire copy)."},{"url":"https://www.nst.com.my/amp/business/corporate/2026/07/1484039/china-plans-let-top-ai-firms-buy-limited-nvidia-h200-chips","date":"2026-07-09","support":"Beijing may allow about 200,000 or fewer NVIDIA H200 chips in total across Alibaba, ByteDance and DeepSeek, so NVIDIA is only a partial fallback. Date form: full date shown (Reuters story)."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://gulfnews.com/technology/huawei-to-double-output-of-top-ai-chip-as-nvidia-wavers-in-china-1.500287469","date":"2025-09-30","support":"Bloomberg: Huawei, the largest domestic supplier, plans about 600,000 910C chips in 2026, roughly double 2025, so spare domestic supply takes one to two years to appear. Date form: full date from the page's last-updated line (Bloomberg report)."},{"url":"https://www.theregister.com/systems/2026/05/22/alibaba-just-admitted-its-struggling-to-keep-up-with-rival-chipmakers-and-ai-shops/5244665","date":"2026-05-22","support":"The Register: Beijing has not let local buyers acquire Nvidia chips, and T-Head can't get much hardware made. Date form: full date shown."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://www.theregister.com/systems/2026/05/22/alibaba-just-admitted-its-struggling-to-keep-up-with-rival-chipmakers-and-ai-shops/5244665","date":"2026-05-22","support":"The Register: Chinese fabs can't yet match TSMC, so domestic capacity is the limit on any replacement. Date form: full date shown."},{"url":"https://gulfnews.com/technology/huawei-to-double-output-of-top-ai-chip-as-nvidia-wavers-in-china-1.500287469","date":"2025-09-30","support":"Bloomberg: SMIC is stuck at 7 nm and Huawei relies on stockpiled foreign HBM, so the best domestic alternative is itself supply-limited. Date form: full date from the page's last-updated line (Bloomberg report)."}]}}}}},"cambricon":{},"cisco":{"a":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"cisco","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"networking-silicon":{"alternatives":{"level":2,"evidence":[{"url":"https://www.delloro.com/news/ai-back-end-networks-continue-their-shift-to-ethernet-now-accounting-for-over-two-thirds-of-3q-2025-switch-sales-in-ai-clusters/","date":"2025-12-05","support":"Celestica and NVIDIA lead AI back-end Ethernet; Accton, Arista and Cisco gained share in 3Q 2025. Full date shown on the source."},{"url":"https://investor.nvidia.com/news/press-release-details/2025/NVIDIA-Spectrum-X-Ethernet-Switches-Speed-Up-Networks-for-Meta-and-Oracle/default.aspx","date":"2025-10-13","support":"Meta also integrates NVIDIA Spectrum switches into FBOSS, alongside other vendors. Full date shown on the source."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://investor.nvidia.com/news/press-release-details/2025/NVIDIA-Spectrum-X-Ethernet-Switches-Speed-Up-Networks-for-Meta-and-Oracle/default.aspx","date":"2025-10-13","support":"Hyperscalers run several switch silicon families under one network OS (FBOSS), so volume can move within months. Full date shown on the source."},{"url":"https://blogs.cisco.com/news/cisco-silicon-one-g200-ai-ml-chip-powers-new-systems-for-hyperscalers-and-enterprises","date":"2024-10-16","support":"Meta planned to deploy Cisco 8501 systems on Silicon One G200 in its network (Cisco blog). Full date shown on the source."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.storagereview.com/news/broadcom-ships-tomahawk-6-switch-series-unleashing-102-4-tbps-switching-capacity","date":"2025-06-09","support":"Broadcom's Tomahawk 6 at 102.4 Tbps doubled the bandwidth of existing Ethernet switches: no gap. Full date shown on the source."},{"url":"https://www.delloro.com/news/ai-back-end-networks-continue-their-shift-to-ethernet-now-accounting-for-over-two-thirds-of-3q-2025-switch-sales-in-ai-clusters/","date":"2025-12-05","support":"Other Ethernet suppliers hold most AI back-end switch sales. Full date shown on the source."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"cisco","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"networking-silicon":{"alternatives":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","date":"2026-02-25","support":"NVIDIA's 10-K lists AMD, Arista, Broadcom, Cisco, HPE, Huawei, Intel, Lumentum and Marvell as networking competitors (Competition). Date form: full filing date from the EDGAR index."},{"url":"https://www.delloro.com/news/ai-back-end-networks-switch-sales-surpass-front-end-networks-for-the-first-time-in-2q2026/","date":"2026-09-03","support":"In 2Q 2026 Celestica leads Ethernet AI back-end switch sales, NVIDIA is a close second, Arista third and Cisco fourth. Date form: full date shown."},{"url":"https://www.theregister.com/networks/2026/06/02/marvell-enters-the-ai-network-fray-with-1024-tbps-switch-silicon/5250180","date":"2026-06-02","support":"Broadcom's Tomahawk 6 is shipping and Marvell's 102.4T T100 is about to sample, so merchant silicon alternatives exist. Date form: full date shown."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.delloro.com/news/ai-back-end-networks-continue-their-shift-to-ethernet-now-accounting-for-over-two-thirds-of-3q-2025-switch-sales-in-ai-clusters/","date":"2025-12-05","support":"Ethernet share of AI back-end switch sales rose from under half to over two-thirds in a year, with Accton, Arista and Cisco gaining the most. Date form: full date shown."},{"url":"https://www.sdxcentral.com/news/ai-hyperscale-and-agentic-demands-helped-cisco-smash-revenue-records/","date":"2026-08-13","support":"Cisco's fiscal 2026 hyperscaler AI orders were $9.3 billion, about 60% Silicon One systems (company statement); competitors' parts are already installed. Date form: full date shown (trade press on Cisco's earnings call)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.delloro.com/news/ai-back-end-networks-switch-sales-surpass-front-end-networks-for-the-first-time-in-2q2026/","date":"2026-09-03","support":"Celestica, NVIDIA and Arista all rank ahead of Cisco in Ethernet AI back-end switch sales (2Q 2026), so the best alternatives already exceed Cisco's scale. Date form: full date shown."},{"url":"https://www.delloro.com/news/ethernet-more-than-doubles-size-of-infiniband-as-the-leading-fabric-for-ai-scale-out-networks-in-2025/","date":"2026-03-10","support":"Celestica and NVIDIA together held about 50 percent of this market in 2025; Cisco was not ranked. Date form: full date shown."}]}}}}},"alchip":{"a":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"alchip","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"custom-accelerators":{"alternatives":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm","date":"2026-03-11","support":"Marvell competes with Alchip and GUC in ASIC design (Item 1, Competition); Broadcom also offers custom accelerators. Full date shown on the source."},{"url":"https://www.techpartner.news/news/google-preparing-to-partner-with-mediatek-on-next-ai-chip-615763","date":"2025-03-18","support":"MediaTek is entering as a TPU design partner, adding another qualified supplier (Reuters). Full date shown on the source."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://www.taipeitimes.com/News/biz/archives/2025/05/30/2003837724","date":"2025-05-30","support":"Alchip's Trainium 3 enters small-volume production at end-2025 and ramps in 2026; replacing it mid-ramp needs a new design cycle. Full date shown on the source."},{"url":"https://investor.marvell.com/news-events/press-releases/detail/121/marvell-expands-strategic-collaboration-with-aws-to-enable-accelerated-infrastructure-for-ai-in-the-cloud","date":"2024-12-02","support":"Marvell holds a five-year, multi-generational custom AI agreement with AWS (company says), the nearest fallback. Full date shown on the source."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://investor.marvell.com/news-events/press-releases/detail/121/marvell-expands-strategic-collaboration-with-aws-to-enable-accelerated-infrastructure-for-ai-in-the-cloud","date":"2024-12-02","support":"Marvell already designs custom AI products for AWS under a multi-generational agreement: a small gap. Full date shown on the source."},{"url":"https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm","date":"2026-03-11","support":"Marvell describes itself as a direct competitor of Alchip in custom silicon (Item 1, Competition). Full date shown on the source."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"alchip","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"custom-accelerators":{"alternatives":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm","date":"2026-03-11","support":"Marvell's 10-K lists Global Unichip, MediaTek, Socionext and Broadcom as direct competitors, with Alchip, so several design partners exist. Date form: full filing date from the EDGAR index."},{"url":"https://counterpointresearch.com/de/insights/AI-Server-Compute-ASIC-Shipments-to-Triple-by-2027","date":"2026-01","support":"Counterpoint lists Alchip with Broadcom, Marvell and MediaTek as design partners for hyperscalers' in-house silicon. Date form: month only, because the page shows 26 January but its dateline year is inconsistent."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://siliconanalysts.com/analysis/custom-ai-chip-economics-tapeout-speed-vs-cost-2026","date":"2026-08-27","support":"Conventional timeline: 18 to 36 months from architecture freeze to first silicon, plus 6 to 12 months of bring-up. Date form: full date shown."},{"url":"https://www.alchip.com/en/Newsroom/Alchip_2025_Q4_financial_results","date":"2026-03-18","support":"Alchip blames its 2025 revenue dip on 'limited production shipments' and expects growth to resume with more 3nm programmes from Q2 2026 (company release). Date form: full date shown."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://counterpointresearch.com/de/insights/AI-Server-Compute-ASIC-Shipments-to-Triple-by-2027","date":"2026-01","support":"Broadcom is projected at about 60% of design-partner share in 2027 and is the largest alternative. Date form: month only, because the page shows 26 January but its dateline year is inconsistent."},{"url":"https://www.sdxcentral.com/news/broadcom-secures-deal-with-google-for-next-gen-ai-racks-and-custom-tpus/","date":"2026-04-07","support":"MediaTek already has orders for v7e versions of Google's Ironwood TPU and will support packaging for TPU v8x, so another partner is in production. Date form: full date shown."}]}}}}},"cerebras":{},"guc":{"a":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"guc","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"custom-accelerators":{"alternatives":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm","date":"2026-03-11","support":"Marvell names GUC and Alchip among ASIC design competitors (Item 1, Competition); Broadcom and MediaTek also compete. Full date shown on the source."},{"url":"https://www.techpartner.news/news/google-preparing-to-partner-with-mediatek-on-next-ai-chip-615763","date":"2025-03-18","support":"MediaTek is preparing to design TPUs for Google, widening the set of design partners (Reuters). Full date shown on the source."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://www.taipeitimes.com/News/biz/archives/2025/05/30/2003837724","date":"2025-05-30","support":"A comparable design partner's AWS 3nm chip moves from small-volume production to ramp over about a year (analogy). Full date shown on the source."},{"url":"https://www.sec.gov/Archives/edgar/data/1807794/000162828026043303/crdo-20260502.htm","date":"2026-06-15","support":"Chip design cycles from engagement to volume typically run two to three years (Credo, Item 1A; analogy). Full date shown on the source."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.taipeitimes.com/News/biz/archives/2025/05/30/2003837724","date":"2025-05-30","support":"Alchip already runs 3nm AI programmes and secures CoWoS capacity for customers: a small gap. Full date shown on the source."},{"url":"https://investor.marvell.com/news-events/press-releases/detail/121/marvell-expands-strategic-collaboration-with-aws-to-enable-accelerated-infrastructure-for-ai-in-the-cloud","date":"2024-12-02","support":"Marvell offers multi-generational custom AI design at hyperscale (company says). Full date shown on the source."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"chips-chip-design","entityId":"guc","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"custom-accelerators":{"alternatives":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm","date":"2026-03-11","support":"Marvell's 10-K names Alchip, Global Unichip, MediaTek, Socionext and Broadcom among direct competitors, so several design partners exist. Date form: full filing date from the EDGAR index."},{"url":"https://counterpointresearch.com/de/insights/AI-Server-Compute-ASIC-Shipments-to-Triple-by-2027","date":"2026-01","support":"Counterpoint names Broadcom, Marvell, Alchip and MediaTek as the active hyperscaler design partners (GUC is not covered). Date form: month only, because the page shows 26 January but its dateline year is inconsistent."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://siliconanalysts.com/analysis/custom-ai-chip-economics-tapeout-speed-vs-cost-2026","date":"2026-08-27","support":"Conventional timeline: 18 to 36 months from architecture freeze to first silicon, plus 6 to 12 months of bring-up. Date form: full date shown."},{"url":"https://futurumgroup.com/insights/jalapeo-in-nine-months-did-ai-just-break-chip-design-timelines/","date":"2026-06-26","support":"Concept to tape-out in 18 to 24 months 'is considered fast' for a frontier accelerator. Date form: full date shown."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://counterpointresearch.com/de/insights/AI-Server-Compute-ASIC-Shipments-to-Triple-by-2027","date":"2026-01","support":"Broadcom is projected at about 60% design-partner share in 2027; Alchip and MediaTek are also in volume. Date form: month only, because the page shows 26 January but its dateline year is inconsistent."},{"url":"https://www.alchip.com/en/Newsroom/Alchip_2025_Q4_financial_results","date":"2026-03-18","support":"Alchip says its 3nm programmes increase from Q2 2026, so a close peer has capacity coming (company release). Date form: full date shown."}]}}}}},"groq":{},"microsoft":{},"tencent":{}},"validation":[{"id":"chips-chip-design-nvidia-accelerator-allocation-2023","key":"chips-chip-design","entityId":"nvidia","description":"In 2023–24, demand for NVIDIA's data-centre GPUs (H100, then H200 and Blackwell) exceeded supply. NVIDIA said it was 'significantly increasing' supply to meet surging demand (May 2023) and remained supply-constrained on H200 and Blackwell (May 2024). Server makers reported GPU lead times of 20+ weeks and lost sales.","sources":[{"url":"https://investor.nvidia.com/news/press-release-details/2023/NVIDIA-Announces-Financial-Results-for-First-Quarter-Fiscal-2024/","title":"NVIDIA Announces Financial Results for First Quarter Fiscal 2024","publisher":"NVIDIA","publishedDate":"2023-05-24","accessedDate":"2026-10-22","tier":"T2","locator":"Jensen Huang: 'We are significantly increasing our supply to meet surging demand for them' (H100 and the data-centre product family)."},{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581024000124/nvda-20240428.htm","title":"NVIDIA Form 10-Q for the quarter ended April 28, 2024","publisher":"NVIDIA / U.S. SEC","publishedDate":"2024-05-29","accessedDate":"2026-10-22","tier":"T1","locator":"MD&A, Recent Developments, Future Objectives and Challenges (Demand and Supply): 'While supply for H100 continued to improve, we are still constrained on H200.' 'We expect supply constraints for our Blackwell offerings will continue into next year.'"},{"url":"https://www.theregister.com/2024/03/01/hpe_q1_2024/","title":"HPE blames GPU shortage for unexpected sales slide","publisher":"The Register (reporting HPE CEO Antonio Neri's earnings remarks)","publishedDate":"2024-03-01","accessedDate":"2026-10-22","tier":"T3","locator":"Neri: HPE 'did not have the GPU supply we wanted'; 'GPU availability remains tight'; GPU lead times at least '20-plus weeks' for H100 and earlier generations."}]}]}