{"schema":"5lc-index-ledger/1","key":"infrastructure-cloud","layer":"infrastructure","sublayer":"cloud","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":"aws","slot":"importance","since":"2026-10-22"},{"entityId":"google-cloud","slot":"importance","since":"2026-10-22"},{"entityId":"oracle-cloud-infrastructure","slot":"importance","since":"2026-10-22"},{"entityId":"microsoft-intelligent-cloud","slot":"importance","since":"2026-10-22"},{"entityId":"coreweave","slot":"importance","since":"2026-10-22"},{"entityId":"alibaba-cloud","slot":"importance","since":"2026-10-22"},{"entityId":"volcano-engine","slot":"importance","since":"2026-10-22"},{"entityId":"nebius","slot":"importance","since":"2026-10-22"},{"entityId":"lambda","slot":"importance","since":"2026-10-22"},{"entityId":"baidu-ai-cloud","slot":"importance","since":"2026-10-22"}],"pending":{},"lastRun":"2026-10-22"},"card":{"schema":"5lc-index-card/1","key":"infrastructure-cloud","layer":"infrastructure","sublayer":"cloud","version":1,"scope":"AI computing capacity that providers rent to customers as a service (GPU and other accelerator instances, clusters and reserved capacity), whether the provider owns its data centres or leases them; leased buildings are scored in Campuses with their operator, servers in Servers & systems, and model APIs sold per token and dedicated model deployments in Models · Inference serving, so each layer of the same capacity is counted once, and computing a company keeps for its own products is not counted here.","outputMeasure":"AI computing rented to customers as a service, measured by revenue and weighted by AI server spending over the latest four quarters.","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","hardToReplace":"GPU clusters take a year or more to build and are reserved for years, so lost capacity cannot quickly move elsewhere. A cloud's own chips are offered mainly by that cloud.","functions":[{"id":"gpu-capacity","name":"GPU computing outside mainland China","why":"Rented GPU clusters carry much of AI training and inference; GPU servers outside China are about three-quarters of AI server spending.","weight":{"low":0.66,"central":0.74,"high":0.87},"weightSource":{"url":"https://www.idc.com/resource-center/blog/ai-infrastructure-spending-holds-near-90-billion-in-q1-2026-as-arm-overtakes-x86-in-accelerated-servers-2026-forecast-raised-to-497-billion/","title":"AI Infrastructure Spending Holds Near $90 Billion in Q1 2026 as ARM Overtakes x86 in Accelerated Servers; 2026 Forecast Raised to $497 Billion","publisher":"IDC","publishedDate":"2026-07-21","accessedDate":"2026-10-22","tier":"T4","locator":"Accelerated (AI) servers by quarter: 3Q25 $81.7B (ARM $29.8B + x86 $51.9B) and 4Q25 $90.2B ($47.5B + $42.7B) from this release; 1Q26 GPU $68.9B and other $17.7B (IDC, 15 Jun 2026); 2Q26 GPU $87.4B and other $27.5B (IDC, 'Worldwide Server Market Revenue Hits Record $166.3 Billion in the Second Quarter of 2026', 10 Sep 2026). 3Q25 and 4Q25 split by interpolating the other-accelerated share between 2Q25 (10.7%) and 1Q26 (20.4%). Four quarters: GPU $301.3B, other $72.1B, total $373.4B; China's 9% removed at 70% GPU and 30% other: GPU outside China $277.8B, 74%. High end allows for the older installed fleet, which has fewer custom-chip servers."},"shareData":"Company GPU rental revenue outside China over the total (company results, Synergy Research)"},{"id":"custom-accelerator-capacity","name":"Computing on clouds' own AI chips","why":"Capacity on chips offered mainly by the cloud that designed them, such as Google TPUs and AWS Trainium; about a sixth of spending.","weight":{"low":0.06,"central":0.17,"high":0.21},"weightSource":{"url":"https://www.idc.com/resource-center/press-releases/1q26-server-tracker/","title":"Worldwide Server Market Revenue Surpasses $122 Billion in the First Quarter of 2026, Driven by AI Infrastructure Demand, according to IDC","publisher":"IDC","publishedDate":"2026-06-15","accessedDate":"2026-10-22","tier":"T4","locator":"'Other Accelerated servers surged 122.1% YoY to $17.7 billion' (1Q26); 2Q26 $27.5B (+237.6%) from IDC's 10 Sep 2026 release; 3Q25 and 4Q25 interpolated ($11.4B and $15.5B). Four quarters $72.1B less China's $10.1B: $62.0B of $373.4B, 17%. Low end 0.06 uses the first-half 2025 mix (8% outside China) for the older installed fleet and allows for custom chips used only inside their owners' products."},"shareData":"Company revenue from renting its own chips outside China over the total (company statements)"},{"id":"china","name":"AI computing rented in mainland China","why":"A separate market under export controls, on imported and domestic chips; about a tenth of spending.","weight":{"low":0.07,"central":0.09,"high":0.12},"weightSource":{"url":"https://www.idc.com/resource-center/blog/ai-infrastructure-spending-holds-near-90-billion-in-q1-2026-as-arm-overtakes-x86-in-accelerated-servers-2026-forecast-raised-to-497-billion/","title":"AI Infrastructure Spending Holds Near $90 Billion in Q1 2026 as ARM Overtakes x86 in Accelerated Servers; 2026 Forecast Raised to $497 Billion","publisher":"IDC","publishedDate":"2026-07-21","accessedDate":"2026-10-22","tier":"T4","locator":"'China (PRC) reached $7.8 billion, an 8.7% share' of AI infrastructure in 1Q26. Matched first-half 2025 check: IDC China's accelerated servers 'US$16.0B' (9 Oct 2025) against worldwide accelerated servers of $139.6B (1Q25 $63.2B and 2Q25 $76.4B, from the growth rates in IDC's 1Q26 and 2Q26 releases), 11.5%. Central 9% for 3Q25–2Q26, between the two; range 7–12%."},"shareData":"Company share of China's AI cloud market (Omdia, IDC China)"}],"risingFloorPct":1,"validationCaseIds":[],"knownGaps":["Weights use AI server spending over four quarters (IDC), a proxy for the installed fleet that earns rental revenue; the older fleet has fewer custom-chip servers.","IDC does not split 3Q25 and 4Q25 AI servers by chip type; those two quarters are interpolated between 2Q25 and 1Q26.","IDC's 'other accelerated' servers include custom chips used only inside their owners' products, which overstates capacity on custom chips that customers can rent.","Hyperscalers report AI revenue as run-rates that include model APIs and some applications, and none reports rental revenue by chip type; their shares rest on estimates.","China's share comes from IDC's AI infrastructure total for 1Q26 and a matched first-half 2025 comparison, and its GPU and non-GPU split from IDC China's accelerator mix.","Private GPU clouds (Lambda, Crusoe, Fluidstack, Together) report revenue rarely; Synergy Research's neocloud total is the main cross-check.","Installed accelerators and contracted power (MW) are reported by specialist clouds only, so they serve as off-basis cross-checks."],"approval":{"approver":"Claude (delegated by Dachi)","date":"2026-10-22","sha256":"5c4582add89f16fcfbdbf2789d748d8fe638ac010efc02c5fa06f1f2830af73b","delegatedOn":"2026-10-22"}},"result":{"schema":"5lc-index-result/1","key":"infrastructure-cloud","asOf":"2026-10-22","methodVersion":"1.1","methodSha256":"3cabed6b38514c263d8b86a494e8cc19cdbd0b2a1013543dc024192c49f7a6a6","cardSha256":"5c4582add89f16fcfbdbf2789d748d8fe638ac010efc02c5fa06f1f2830af73b","validationSha256":"409931bb458f38b02342c9bf43e61eab299e47d8679912af2c96074ec556f17e","inputsSha256":"de870252cf50233e8ba760f8e0fc7de67196846b2f165ef7f1c8008a9a997a50","unscreened":[],"companies":[{"entityId":"aws","name":"Amazon Web Services (AWS)","reachPct":17.02437,"importance":14.4558972282,"importanceRank":1,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":12.768,"shareLowPct":5.1072,"shareHighPct":20.4288,"stale":false,"offBasis":false,"level":2.33333333333,"years":0.6299605249459811,"disagreement":false},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":44.565,"shareLowPct":17.826,"shareHighPct":71.304,"stale":false,"offBasis":false,"level":3.16666666667,"years":1.1224620483119663,"disagreement":false},{"id":"china","weight":0.09,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{"committed":0},"mc":{"medianRank":2,"rankP10":1,"rankP90":4,"pTop8":1,"pSelected":1,"stability":"firm"},"baselineRevenueRank":1},{"entityId":"google-cloud","name":"Google Cloud","reachPct":15.30683,"importance":14.4153110971,"importanceRank":2,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":10.447,"shareLowPct":4.1788,"shareHighPct":16.7152,"stale":false,"offBasis":false,"level":2.33333333333,"years":0.6299605249459811,"disagreement":false},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":44.565,"shareLowPct":17.826,"shareHighPct":71.304,"stale":false,"offBasis":false,"level":3.33333333333,"years":1.2599210498919622,"disagreement":false},{"id":"china","weight":0.09,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{"committed":25},"mc":{"medianRank":2,"rankP10":1,"rankP90":4,"pTop8":1,"pSelected":1,"stability":"firm"},"baselineRevenueRank":3},{"entityId":"oracle-cloud-infrastructure","name":"Oracle Cloud Infrastructure (OCI)","reachPct":17.1791,"importance":12.1474581047,"importanceRank":3,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":23.215,"shareLowPct":9.286,"shareHighPct":37.144,"stale":false,"offBasis":false,"level":2.5,"years":0.7071067811865475,"disagreement":false},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":91.6666666667,"risingEligible":true,"risingComponents":{"growth":75,"shareChange":100,"committed":100},"mc":{"medianRank":3,"rankP10":1,"rankP90":4,"pTop8":1,"pSelected":1,"stability":"firm"},"baselineRevenueRank":6},{"entityId":"microsoft-intelligent-cloud","name":"Microsoft Azure","reachPct":15.46082,"importance":10.9324506647,"importanceRank":4,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":20.893,"shareLowPct":8.3572,"shareHighPct":33.4288,"stale":false,"offBasis":false,"level":2.5,"years":0.7071067811865475,"disagreement":false},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":62.5,"risingEligible":true,"risingComponents":{"growth":50,"shareChange":75},"mc":{"medianRank":3,"rankP10":1,"rankP90":4,"pTop8":1,"pSelected":1,"stability":"firm"},"baselineRevenueRank":2},{"entityId":"coreweave","name":"CoreWeave","reachPct":8.84744,"importance":5.57353794683,"importanceRank":5,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":11.956,"shareLowPct":4.7824,"shareHighPct":19.1296,"stale":false,"offBasis":false,"level":2.33333333333,"years":0.6299605249459811,"disagreement":false},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":25,"risingEligible":true,"risingComponents":{"growth":0,"shareChange":25,"committed":50},"mc":{"medianRank":5,"rankP10":4,"rankP90":6,"pTop8":1,"pSelected":1,"stability":"firm"},"baselineRevenueRank":7},{"entityId":"alibaba-cloud","name":"Alibaba Cloud","reachPct":3.77236,"importance":3.1407317045,"importanceRank":6,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.464,"shareLowPct":0.1856,"shareHighPct":0.7424,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":38.1,"shareLowPct":13.335,"shareHighPct":62.865,"stale":false,"offBasis":true,"level":2.83333333333,"years":0.890898718138281,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{},"mc":{"medianRank":6,"rankP10":5,"rankP90":6,"pTop8":1,"pSelected":1,"stability":"firm"},"baselineRevenueRank":4},{"entityId":"volcano-engine","name":"Volcano Engine (ByteDance)","reachPct":1.92184,"importance":1.31970805026,"importanceRank":7,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.116,"shareLowPct":0.0464,"shareHighPct":0.1856,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":20.4,"shareLowPct":7.14,"shareHighPct":33.66,"stale":false,"offBasis":true,"level":2.5,"years":0.7071067811865475,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{},"mc":{"medianRank":7,"rankP10":7,"rankP90":8,"pTop8":0.9918,"pSelected":0.9999,"stability":"firm"},"baselineRevenueRank":null},{"entityId":"nebius","name":"Nebius","reachPct":1.9758,"importance":0.784096749622,"importanceRank":8,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":2.67,"shareLowPct":1.068,"shareHighPct":4.272,"stale":false,"offBasis":false,"level":1.66666666667,"years":0.3968502629929668,"disagreement":false},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":75,"risingEligible":true,"risingComponents":{"growth":100,"shareChange":50,"committed":75},"mc":{"medianRank":8,"rankP10":7,"rankP90":9,"pTop8":0.8979,"pSelected":1,"stability":"firm"},"baselineRevenueRank":10},{"entityId":"lambda","name":"Lambda","reachPct":1.28834,"importance":0.405801671354,"importanceRank":9,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":1.741,"shareLowPct":0.6964,"shareHighPct":2.7856,"stale":false,"offBasis":false,"level":1.33333333333,"years":0.3149802624729905,"disagreement":false},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":12.5,"risingEligible":true,"risingComponents":{"growth":25,"shareChange":0},"mc":{"medianRank":9,"rankP10":9,"rankP90":12,"pTop8":0.0631,"pSelected":0.7699,"stability":"likely"},"baselineRevenueRank":13},{"entityId":"baidu-ai-cloud","name":"Baidu AI Cloud","reachPct":0.86302,"importance":0.339990322492,"importanceRank":10,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.023,"shareLowPct":0.0092,"shareHighPct":0.0368,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":9.4,"shareLowPct":3.29,"shareHighPct":15.51,"stale":false,"offBasis":true,"level":1.66666666667,"years":0.3968502629929668,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":10,"rankP10":9,"rankP90":14,"pTop8":0.0389,"pSelected":0.5525,"stability":"borderline"},"baselineRevenueRank":9},{"entityId":"iren","name":"IREN","reachPct":0.85914,"importance":0.270612142701,"importanceRank":11,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":1.161,"shareLowPct":0.327,"shareHighPct":1.857,"stale":false,"offBasis":false,"level":1.33333333333,"years":0.3149802624729905,"disagreement":false},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":12,"rankP10":10,"rankP90":16,"pTop8":0.0023,"pSelected":0.2087,"stability":"borderline"},"baselineRevenueRank":16},{"entityId":"tencent-cloud","name":"Tencent Cloud","reachPct":0.69576,"importance":0.257204099117,"importanceRank":12,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.174,"shareLowPct":0.0696,"shareHighPct":0.2784,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":6.3,"shareLowPct":2.205,"shareHighPct":10.395,"stale":false,"offBasis":true,"level":1.66666666667,"years":0.3968502629929668,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":12,"rankP10":10,"rankP90":15,"pTop8":0.0034,"pSelected":0.2219,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"huawei-cloud","name":"Huawei Cloud","reachPct":0.58315,"importance":0.224944490423,"importanceRank":13,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.116,"shareLowPct":0.0464,"shareHighPct":0.1856,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0.543,"shareLowPct":0.109,"shareHighPct":1.63,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"china","weight":0.09,"sharePct":4.5,"shareLowPct":0.875,"shareHighPct":7.025,"stale":false,"offBasis":true,"level":1.83333333333,"years":0.4454493590691404,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":13,"rankP10":11,"rankP90":17,"pTop8":0.0007,"pSelected":0.0692,"stability":"borderline"},"baselineRevenueRank":8},{"entityId":"china-telecom-tianyi-cloud","name":"China Telecom (Tianyi Cloud)","reachPct":0.56504,"importance":0.219237489649,"importanceRank":14,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.046,"shareLowPct":0.0184,"shareHighPct":0.0736,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":5.9,"shareLowPct":2.065,"shareHighPct":9.735,"stale":false,"offBasis":true,"level":1.66666666667,"years":0.3968502629929668,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":13,"rankP10":10,"rankP90":18,"pTop8":0.0013,"pSelected":0.1055,"stability":"borderline"},"baselineRevenueRank":5},{"entityId":"crusoe","name":"Crusoe","reachPct":0.68746,"importance":0.171865,"importanceRank":15,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.929,"shareLowPct":0.3716,"shareHighPct":1.4864,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":14,"rankP10":11,"rankP90":18,"pTop8":0.0006,"pSelected":0.0595,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"together-ai","name":"Together AI","reachPct":0.55796,"importance":0.13949,"importanceRank":16,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.754,"shareLowPct":0.3016,"shareHighPct":1.2064,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":16,"rankP10":13,"rankP90":20,"pTop8":0,"pSelected":0.0113,"stability":"borderline"},"baselineRevenueRank":12},{"entityId":"kingsoft-cloud","name":"Kingsoft Cloud","reachPct":0.43904,"importance":0.136077006302,"importanceRank":17,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.046,"shareLowPct":0.0184,"shareHighPct":0.0736,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":4.5,"shareLowPct":1.875,"shareHighPct":7.025,"stale":false,"offBasis":true,"level":1.33333333333,"years":0.3149802624729905,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":18,"rankP10":15,"rankP90":21,"pTop8":0,"pSelected":0.0004,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"core42","name":"Core42 (G42)","reachPct":0.4292,"importance":0.1073,"importanceRank":18,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.58,"shareLowPct":0.232,"shareHighPct":0.928,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"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":15,"rankP90":23,"pTop8":0,"pSelected":0.0006,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"china-mobile-cloud","name":"China Mobile Cloud","reachPct":0.2959,"importance":0.10193441546,"importanceRank":19,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.035,"shareLowPct":0.014,"shareHighPct":0.056,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":3,"shareLowPct":0.45,"shareHighPct":5.55,"stale":false,"offBasis":true,"level":1.5,"years":0.3535533905932738,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":21,"rankP10":17,"rankP90":28,"pTop8":0,"pSelected":0.0001,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"fluidstack","name":"Fluidstack","reachPct":0.38628,"importance":0.09657,"importanceRank":20,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.522,"shareLowPct":0.2088,"shareHighPct":0.8352,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"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":16,"rankP90":24,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"modal","name":"Modal","reachPct":0.38628,"importance":0.09657,"importanceRank":21,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.522,"shareLowPct":0.2088,"shareHighPct":0.8352,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"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":16,"rankP90":24,"pTop8":0,"pSelected":0.0002,"stability":"borderline"},"baselineRevenueRank":11},{"entityId":"voltage-park","name":"Voltage Park","reachPct":0.25752,"importance":0.06438,"importanceRank":22,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.348,"shareLowPct":0.1392,"shareHighPct":0.5568,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"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":20,"rankP90":29,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"nscale","name":"Nscale","reachPct":0.2405,"importance":0.060125,"importanceRank":23,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.325,"shareLowPct":0.13,"shareHighPct":0.52,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":25,"rankP10":20,"rankP90":30,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":14},{"entityId":"china-unicom-cloud","name":"China Unicom Cloud","reachPct":0.17902,"importance":0.0552818025206,"importanceRank":24,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.023,"shareLowPct":0.0092,"shareHighPct":0.0368,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":1.8,"shareLowPct":0.27,"shareHighPct":3.33,"stale":false,"offBasis":true,"level":1.33333333333,"years":0.3149802624729905,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":28,"rankP10":23,"rankP90":35,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"runpod","name":"Runpod","reachPct":0.20646,"importance":0.051615,"importanceRank":25,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.279,"shareLowPct":0.1116,"shareHighPct":0.4464,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":27,"rankP10":22,"rankP90":32,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":15},{"entityId":"sensetime","name":"SenseTime (SenseCore)","reachPct":0.1609,"importance":0.0489973354339,"importanceRank":26,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.035,"shareLowPct":0.014,"shareHighPct":0.056,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":1.5,"shareLowPct":0.225,"shareHighPct":2.775,"stale":false,"offBasis":true,"level":1.33333333333,"years":0.3149802624729905,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":29,"rankP10":24,"rankP90":35,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"nvidia","name":"NVIDIA (DGX Cloud)","reachPct":0.17168,"importance":0.04292,"importanceRank":27,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.232,"shareLowPct":0.058,"shareHighPct":0.929,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"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":17,"rankP90":32,"pTop8":0,"pSelected":0.0003,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"rumble","name":"Rumble (Northern Data, Taiga Cloud)","reachPct":0.17168,"importance":0.04292,"importanceRank":28,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.232,"shareLowPct":0.0928,"shareHighPct":0.3712,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":29,"rankP10":24,"rankP90":34,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"sakura-internet","name":"Sakura Internet","reachPct":0.0999,"importance":0.024975,"importanceRank":29,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.135,"shareLowPct":0.054,"shareHighPct":0.216,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":35,"rankP10":31,"rankP90":38,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"mistral","name":"Mistral AI (Mistral Compute)","reachPct":0.08584,"importance":0.02146,"importanceRank":30,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.116,"shareLowPct":0.029,"shareHighPct":0.464,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":32,"rankP10":24,"rankP90":38,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"naver-cloud","name":"NAVER Cloud","reachPct":0.08584,"importance":0.02146,"importanceRank":31,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.116,"shareLowPct":0.029,"shareHighPct":0.464,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":31,"rankP10":23,"rankP90":38,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"tensorwave","name":"TensorWave","reachPct":0.08584,"importance":0.02146,"importanceRank":32,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.116,"shareLowPct":0.029,"shareHighPct":0.464,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":32,"rankP10":24,"rankP90":38,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"vultr","name":"Vultr","reachPct":0.08584,"importance":0.02146,"importanceRank":33,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.116,"shareLowPct":0.029,"shareHighPct":0.464,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":32,"rankP10":24,"rankP90":38,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"gmo-internet","name":"GMO Internet","reachPct":0.06882,"importance":0.017205,"importanceRank":34,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.093,"shareLowPct":0.023,"shareHighPct":0.371,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":34,"rankP10":26,"rankP90":38,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"yotta","name":"Yotta Data Services (Shakti Cloud)","reachPct":0.06882,"importance":0.017205,"importanceRank":35,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.093,"shareLowPct":0.023,"shareHighPct":0.371,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":34,"rankP10":27,"rankP90":38,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"nexgen-cloud","name":"NexGen Cloud (Hyperstack)","reachPct":0.05994,"importance":0.014985,"importanceRank":36,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.081,"shareLowPct":0.0648,"shareHighPct":0.5042,"stale":true,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":31,"rankP10":23,"rankP90":37,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"softbank-corp","name":"SoftBank Corp.","reachPct":0.0481,"importance":0.012025,"importanceRank":37,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.065,"shareLowPct":0.013,"shareHighPct":0.427,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":34,"rankP10":25,"rankP90":39,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"kddi","name":"KDDI (GPU cloud)","reachPct":0.03922,"importance":0.009805,"importanceRank":38,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.053,"shareLowPct":0.011,"shareHighPct":0.427,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":34,"rankP10":25,"rankP90":39,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"ntt","name":"NTT (GPU cloud services)","reachPct":0.03256,"importance":0.00814,"importanceRank":39,"functions":[{"id":"gpu-capacity","weight":0.74,"sharePct":0.044,"shareLowPct":0.009,"shareHighPct":0.35,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"custom-accelerator-capacity","weight":0.17,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china","weight":0.09,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":36,"rankP10":28,"rankP90":39,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null}],"selection":{"importanceIds":["aws","google-cloud","oracle-cloud-infrastructure","microsoft-intelligent-cloud","coreweave","alibaba-cloud","volcano-engine","nebius"],"risingIds":[],"filledFromImportance":["lambda","baidu-ai-cloud"]},"risingPools":{"growth":5,"shareChange":5,"committed":5},"weightSensitivity":[{"functionId":"gpu-capacity","end":"low","entered":[],"left":[]},{"functionId":"gpu-capacity","end":"high","entered":["iren"],"left":["baidu-ai-cloud"]},{"functionId":"custom-accelerator-capacity","end":"low","entered":[],"left":[]},{"functionId":"custom-accelerator-capacity","end":"high","entered":[],"left":[]},{"functionId":"china","end":"low","entered":["iren"],"left":["baidu-ai-cloud"]},{"functionId":"china","end":"high","entered":[],"left":[]}],"validation":[],"baseline":{"spearman":0.761764705882,"baselineTop10NotSelected":["china-telecom-tianyi-cloud","huawei-cloud"],"selectedNotInBaselineTop10":["volcano-engine","lambda"]},"raterAgreement":{"within":0.75,"weightedKappa":0.683098591549,"disagreements":[]}},"inputs":{"aws":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"aws","name":"Amazon Web Services (AWS)","kind":"unit","parentId":"amazon","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":12.768,"central":12.768,"high":12.768,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"Q2 2026 annualised","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0001018724/000101872426000024/amzn-20260630xex991.htm","title":"Amazon.com Announces Second Quarter Results (Form 8-K exhibit 99.1)","publisher":"Amazon.com, Inc.","publishedDate":"2026-07-30","accessedDate":"2026-10-22","tier":"T5","locator":"AWS net sales Q2 2026 $42,232M (+37%); 'Exceeded a $25 billion annual revenue run rate for AWS's AI business, growing triple-digit percentages year-over-year'; chips business also above $25B"},"note":"5LayerCake T5 estimate (assumed split). The company says AWS's AI business passed a $25B annual revenue run rate (growing triple digits) and its chips business (which includes Graviton and Trainium; the 8-K gives no product breakdown, and the AI and chips figures overlap) passed $25B; AWS does not report GPU rental. Assumed split of the $25B AI run rate: NVIDIA GPU capacity rented about $11B, Trainium capacity about $5B, and about $9B for Bedrock model APIs and tools (per-token APIs belong to Inference serving). 11.0 / 86.2. Denominator (the same in every file of this sublayer): GPU computing outside mainland China = about $86.2B a year, annualised from Q2 2026 (Oracle: June-August 2026). It is the sum of the provider estimates: AWS 11.0, Azure 18.0 and Google Cloud 9.0 (assumed splits of company-reported AI figures; none reports GPU rental), Oracle OCI 20.0, CoreWeave 10.3 (Q2 revenue $2,575M x 4), Nebius 2.3 (AI cloud revenue $574.9M x 4), all other registered providers 9.6, and 6.0 for providers outside the register. Reconciliation with published totals: Gartner forecasts AI-optimized IaaS spending of $42.3B for all of 2026 (Techstrong, 2026-09-21; Gartner's 10 August 2026 release is titled as a worldwide forecast; gartner.com returned HTTP 403 and was not retried, so its definition is unread). CoreWeave 10.3 + Nebius 2.3 + Oracle's CPU and GPU compute revenue of 26.0 ($6.5B in June-August 2026 x 4, The Next Platform, 2026-09-14) already total 38.6, 91% of Gartner's whole-year worldwide figure, before AWS, Microsoft, Google or any other provider. Gartner's category must therefore be narrower than AI compute rented to customers (or low), so it is not used as the denominator; it is the lowest case. Synergy's neocloud revenue above $25B for 2025 (2026-04-02) counts GPU-as-a-service, generative-AI platforms and high-density data-centre capacity and lists OpenAI and Anthropic among cloud providers, so it is wider than rented GPUs; the registered neoclouds' rental revenue sums to about 19.4 here. 5LayerCake judgement of the plausible range for the total: about $58B (every provider at the low end of its own range) to about $115B; the T5 share width covers it. Hyperscalers resell some capacity bought from neoclouds, so the total can double count by a few percent. No growth or prior-share input: Amazon gives the AWS AI run rate only as 'more than $25B', 'growing triple-digit percentages', with no prior-year figure; there is no like-for-like prior."},"custom-accelerator-capacity":{"low":44.565,"central":44.565,"high":44.565,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"Q2 2026 annualised","periodEnd":"2026-06-30","source":{"url":"https://www.anthropic.com/news/anthropic-amazon-compute","title":"Anthropic and Amazon expand compute collaboration","publisher":"Anthropic","publishedDate":"2026-04-20","accessedDate":"2026-10-22","tier":"T5","locator":"Anthropic: 'we currently use over one million Trainium2 chips to train and serve Claude'; 'nearly 1GW total of Trainium2 and Trainium3 capacity coming online by the end of 2026'; 'Significant Trainium2 capacity is coming online in Q2'"},"note":"5LayerCake T5 estimate (assumed): AWS Trainium capacity rented about $4.1B a year (Q2 2026 annualised). Arithmetic per year, on the card's basis (rented computing outside China, Q2 2026 annualised). (1) Revenue per gigawatt-year: CoreWeave's Q2 2026 results (Form 8-K exhibit 99.1, 2026-08-11, T2, https://www.sec.gov/Archives/edgar/data/1769628/000176962826000362/coreweave2q26earningspress.htm) give quarterly revenue of $2,575M and say active power was expanded 'by nearly 500 MWs to reach 1.5 GW', so about 1.0 GW at the start of the quarter and about 1.25 GW on average (linear ramp assumed). $2,575M x 4 = $10.3B; 10.3 / 1.25 = $8.2B per gigawatt-year (range 10.3 / 1.5 = $6.9B to 10.3 / 1.0 = $10.3B). That is a rate for rented GPUs, applied unchanged to custom chips (assuming Anthropic's 'GW' and CoreWeave's 'active power' mean the same load, which neither release defines): AWS says Trn2 offers '30-40% better price performance' than GPU instances (https://aws.amazon.com/ec2/instance-types/trn2/), which could lower dollars per unit of compute, and the Trn2 page shows no price. (2) Gigawatts in service: see the cloud-specific sentence below. (3) 0.5 GW x $8.2B = $4.1B a year (range 0.3 GW x 6.9 = $2.1B to 0.7 GW x 10.3 = $7.2B), the same for each cloud by the labelled judgement below. Gigawatts in service: Anthropic says it uses 'over one million Trainium2 chips' and expects 'nearly 1GW total of Trainium2 and Trainium3 capacity' online by the end of 2026, with significant Trainium2 capacity arriving in Q2 (T2); it gives no mid-year figure and does not state power per chip, so about 0.5 GW is assumed in service on average in Q2 2026, half the year-end target (range 0.3 to 0.7 GW; 5LayerCake judgement). Amazon reports no Trainium revenue: its Q2 2026 results release says the AI business and the chips business each passed a $25B annual run rate and gives no product breakdown (https://www.sec.gov/Archives/edgar/data/0001018724/000101872426000024/amzn-20260630xex991.htm), so neither is used for this function. Denominator: rented computing on cloud-owned AI chips outside China = about $9.2B a year (5LayerCake estimate, Q2 2026 annualised): Google TPU capacity rented 4.1, AWS Trainium capacity rented 4.1, other custom-chip clouds 1.0 (a 5LayerCake judgement; no source sizes it; it includes Huawei Cloud's Ascend allowance of about 0.05, recorded in huawei-cloud.json). 4.1 / 9.2 = 44.565%. The equal Google/AWS split is a labelled 5LayerCake judgement, not a measurement: each cloud has one anchor customer with a similar 2026 target (Anthropic: nearly 1 GW of Trainium2 and Trainium3 by the end of 2026, https://www.anthropic.com/news/anthropic-amazon-compute, 2026-04-20; 'well over a gigawatt' of TPU capacity online in 2026, https://www.anthropic.com/news/expanding-our-use-of-google-cloud-tpus-and-services, 2025-10-23), and no source splits rental revenue by cloud. The share moves little with the level: with each cloud between 2.1 and 7.2 it is 40% to 47%. Rechecked division: an earlier version of this note took a TPU rental rate of $13.1-17.5B per gigawatt 'over three years' from a TipRanks article (a finance portal, no longer used) and divided by three. Those two figures equal 1,000,000 chips x $1.50 and $2.00 an hour x 8,760 hours = $13.14B and $17.52B, which is one year for a million-chip fleet, so the division by three was wrong; the figures are dropped and replaced by the CoreWeave rate above. Google reports TPU system sales as hardware from Q2 2026 (Alphabet 8-K), which are sales, not rental, and are not counted. Anthropic's reported $200B five-year Google Cloud commitment (Reuters via The Standard, 2026-05-06, citing The Information, unconfirmed, https://www.thestandard.com.hk/finance/article/331210/Anthropic-commits-to-spending-US200-billion-on-Googles-cloud-and-chips-the-Information-reports) and the multi-gigawatt TPU capacity Anthropic expects from 2027 (https://www.anthropic.com/news/google-broadcom-partnership-compute, 2026-04-06) start after this period and are context only. Screening used 15.0 for Google and 8.5 for AWS; these figures replace them."},"china":{"participates":false,"evidence":{"url":"https://www.amazonaws.cn/en/about-aws/","title":"About AWS (China)","publisher":"Amazon Web Services China","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"States that Sinnet is 'the service operator and provider' for the Amazon Web Services China (Beijing) Region and NWCD for the China (Ningxia) Region, and that AWS 'has partnered with' them. The page does not say who bills customers, so AWS's own revenue from AI compute rented in mainland China is judged to be nil."}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":3},"custom-accelerator-capacity":{"levelLow":2,"levelHigh":5}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"5399b22a437efe21529162311515e86818f8f6241629d7452d8f1e117e5ff4b4"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"5399b22a437efe21529162311515e86818f8f6241629d7452d8f1e117e5ff4b4"}},"committedExpansion":{"low":2.94,"central":2.94,"high":2.94,"unit":"ratio","basis":"AWS remaining performance obligations ÷ AWS annualised net sales (segment-wide, includes non-AI contracts)","period":"RPO at 30 June 2026","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/1018724/000101872426000026/amzn-20260630.htm","title":"Amazon.com, Inc. Form 10-Q for the quarter ended June 30, 2026","publisher":"Amazon.com, Inc.","publishedDate":"2026-07-31","accessedDate":"2026-10-22","tier":"T2","locator":"Remaining performance obligations: 'approximately $496 billion as of June 30, 2026', 'primarily related to AWS'; weighted-average remaining life 6.4 years (numerator is a T1 filing; denominator is the T2 segment run rate)"},"note":"Form 10-Q (T1): remaining performance obligations about $496B at 30 June 2026, primarily AWS; AWS net sales Q2 2026 $42,232M x 4 = $168.9B (company release, T2); 496 / 168.9 = 2.94. Both are AWS-wide, so non-AI contracts are inside both numbers; the filing names OpenAI ($38B expanded by $100B over 8 years) and Anthropic (expanded by more than $100B over 10 years). Tier is the lower of the two (T2)."},"baselineRevenueUsd":{"low":168928000000,"central":168928000000,"high":168928000000,"unit":"USD","basis":"AWS segment net sales, latest quarter x 4, US dollars","period":"Q2 2026 annualised","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0001018724/000101872426000024/amzn-20260630xex991.htm","title":"Amazon.com Announces Second Quarter Results (Form 8-K exhibit 99.1)","publisher":"Amazon.com, Inc.","publishedDate":"2026-07-30","accessedDate":"2026-10-22","tier":"T2","locator":"Segment table: AWS net sales Q2 2026 $42,232M; 'a $169 billion annualized revenue run rate'"},"note":"Segment net sales of the unit (never Amazon's total). Annualised from the latest quarter, as the company does."}},"google-cloud":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"google-cloud","name":"Google Cloud","kind":"unit","parentId":"alphabet","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":10.447,"central":10.447,"high":10.447,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"Q2 2026 annualised","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0001652044/000165204426000066/googexhibit991q22026.htm","title":"Alphabet Announces Second Quarter 2026 Results (Form 8-K exhibit 99.1)","publisher":"Alphabet Inc.","publishedDate":"2026-07-22","accessedDate":"2026-10-22","tier":"T5","locator":"Google Cloud revenue Q2 2026 $24,768M (Q2 2025 $13,624M, +82%); growth 'led by an increase in Google Cloud Platform (GCP) across enterprise AI Solutions and enterprise AI Infrastructure'; CEO remarks: accelerators from Google and NVIDIA"},"note":"5LayerCake T5 estimate (assumed split). Google Cloud revenue rose from $13.6B to $24.8B a quarter; the company attributes growth to AI infrastructure and AI solutions and does not split them or report GPU rental. Assumed NVIDIA GPU capacity rented about $9B a year (about 9% of the $99B annualised segment). 9.0 / 86.2. Denominator (the same in every file of this sublayer): GPU computing outside mainland China = about $86.2B a year, annualised from Q2 2026 (Oracle: June-August 2026). It is the sum of the provider estimates: AWS 11.0, Azure 18.0 and Google Cloud 9.0 (assumed splits of company-reported AI figures; none reports GPU rental), Oracle OCI 20.0, CoreWeave 10.3 (Q2 revenue $2,575M x 4), Nebius 2.3 (AI cloud revenue $574.9M x 4), all other registered providers 9.6, and 6.0 for providers outside the register. Reconciliation with published totals: Gartner forecasts AI-optimized IaaS spending of $42.3B for all of 2026 (Techstrong, 2026-09-21; Gartner's 10 August 2026 release is titled as a worldwide forecast; gartner.com returned HTTP 403 and was not retried, so its definition is unread). CoreWeave 10.3 + Nebius 2.3 + Oracle's CPU and GPU compute revenue of 26.0 ($6.5B in June-August 2026 x 4, The Next Platform, 2026-09-14) already total 38.6, 91% of Gartner's whole-year worldwide figure, before AWS, Microsoft, Google or any other provider. Gartner's category must therefore be narrower than AI compute rented to customers (or low), so it is not used as the denominator; it is the lowest case. Synergy's neocloud revenue above $25B for 2025 (2026-04-02) counts GPU-as-a-service, generative-AI platforms and high-density data-centre capacity and lists OpenAI and Anthropic among cloud providers, so it is wider than rented GPUs; the registered neoclouds' rental revenue sums to about 19.4 here. 5LayerCake judgement of the plausible range for the total: about $58B (every provider at the low end of its own range) to about $115B; the T5 share width covers it. Hyperscalers resell some capacity bought from neoclouds, so the total can double count by a few percent. No growth or prior-share input: Alphabet does not report AI revenue or GPU/TPU rental revenue, so no like-for-like prior exists."},"custom-accelerator-capacity":{"low":44.565,"central":44.565,"high":44.565,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"Q2 2026 annualised","periodEnd":"2026-06-30","source":{"url":"https://www.anthropic.com/news/expanding-our-use-of-google-cloud-tpus-and-services","title":"Expanding our use of Google Cloud TPUs","publisher":"Anthropic","publishedDate":"2025-10-23","accessedDate":"2026-10-22","tier":"T5","locator":"Anthropic: 'up to one million TPUs'; 'The expansion is worth tens of billions of dollars'; 'expected to bring well over a gigawatt of capacity online in 2026'"},"note":"5LayerCake T5 estimate (assumed): Google TPU capacity rented about $4.1B a year (Q2 2026 annualised). Arithmetic per year, on the card's basis (rented computing outside China, Q2 2026 annualised). (1) Revenue per gigawatt-year: CoreWeave's Q2 2026 results (Form 8-K exhibit 99.1, 2026-08-11, T2, https://www.sec.gov/Archives/edgar/data/1769628/000176962826000362/coreweave2q26earningspress.htm) give quarterly revenue of $2,575M and say active power was expanded 'by nearly 500 MWs to reach 1.5 GW', so about 1.0 GW at the start of the quarter and about 1.25 GW on average (linear ramp assumed). $2,575M x 4 = $10.3B; 10.3 / 1.25 = $8.2B per gigawatt-year (range 10.3 / 1.5 = $6.9B to 10.3 / 1.0 = $10.3B). That is a rate for rented GPUs, applied unchanged to custom chips (assuming Anthropic's 'GW' and CoreWeave's 'active power' mean the same load, which neither release defines): AWS says Trn2 offers '30-40% better price performance' than GPU instances (https://aws.amazon.com/ec2/instance-types/trn2/), which could lower dollars per unit of compute, and the Trn2 page shows no price. (2) Gigawatts in service: see the cloud-specific sentence below. (3) 0.5 GW x $8.2B = $4.1B a year (range 0.3 GW x 6.9 = $2.1B to 0.7 GW x 10.3 = $7.2B), the same for each cloud by the labelled judgement below. Gigawatts in service: Anthropic says its expansion with Google Cloud covers 'up to one million TPUs' and 'well over a gigawatt of capacity' online in 2026 (T2); Alphabet gives no TPU capacity figure and says Google Cloud product revenue now comes 'primarily from the sale of TPU systems' (Q2 2026 8-K, https://www.sec.gov/Archives/edgar/data/0001652044/000165204426000066/googexhibit991q22026.htm; sales, not rental, so not counted here). About 0.5 GW is assumed in service on average in Q2 2026, half of a year-end level just above 1 GW (range 0.3 to 0.7 GW; 5LayerCake judgement, no mid-year figure exists), plus a few other TPU renters inside the same figure. For context, Google's Cloud TPU price list (https://cloud.google.com/tpu/pricing, opened 2026-10-22) shows Ironwood at $12.00 per chip-hour on demand and $5.40 with a three-year commitment (us-central1): at the three-year rate a million Ironwood chips would be $47.3B a year, far above this estimate, so list prices are not used; large buyers pay negotiated rates that are not disclosed. Denominator: rented computing on cloud-owned AI chips outside China = about $9.2B a year (5LayerCake estimate, Q2 2026 annualised): Google TPU capacity rented 4.1, AWS Trainium capacity rented 4.1, other custom-chip clouds 1.0 (a 5LayerCake judgement; no source sizes it; it includes Huawei Cloud's Ascend allowance of about 0.05, recorded in huawei-cloud.json). 4.1 / 9.2 = 44.565%. The equal Google/AWS split is a labelled 5LayerCake judgement, not a measurement: each cloud has one anchor customer with a similar 2026 target (Anthropic: nearly 1 GW of Trainium2 and Trainium3 by the end of 2026, https://www.anthropic.com/news/anthropic-amazon-compute, 2026-04-20; 'well over a gigawatt' of TPU capacity online in 2026, https://www.anthropic.com/news/expanding-our-use-of-google-cloud-tpus-and-services, 2025-10-23), and no source splits rental revenue by cloud. The share moves little with the level: with each cloud between 2.1 and 7.2 it is 40% to 47%. Rechecked division: an earlier version of this note took a TPU rental rate of $13.1-17.5B per gigawatt 'over three years' from a TipRanks article (a finance portal, no longer used) and divided by three. Those two figures equal 1,000,000 chips x $1.50 and $2.00 an hour x 8,760 hours = $13.14B and $17.52B, which is one year for a million-chip fleet, so the division by three was wrong; the figures are dropped and replaced by the CoreWeave rate above. Google reports TPU system sales as hardware from Q2 2026 (Alphabet 8-K), which are sales, not rental, and are not counted. Anthropic's reported $200B five-year Google Cloud commitment (Reuters via The Standard, 2026-05-06, citing The Information, unconfirmed, https://www.thestandard.com.hk/finance/article/331210/Anthropic-commits-to-spending-US200-billion-on-Googles-cloud-and-chips-the-Information-reports) and the multi-gigawatt TPU capacity Anthropic expects from 2027 (https://www.anthropic.com/news/google-broadcom-partnership-compute, 2026-04-06) start after this period and are context only. Screening used 15.0 for Google and 8.5 for AWS; these figures replace them."},"china":{"participates":false,"evidence":{"url":"https://cloud.google.com/about/locations","title":"Cloud locations","publisher":"Google Cloud","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Cloud locations page (43 regions, 130 zones as retrieved) names no mainland China region; the retrieved text names only Osaka among Asian regions"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":3},"custom-accelerator-capacity":{"levelLow":2,"levelHigh":5}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"a94d21ae25d13a84d2daebae9957a90a2eb4c679d04ed1f403a2ef801944782d"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"a94d21ae25d13a84d2daebae9957a90a2eb4c679d04ed1f403a2ef801944782d"}},"committedExpansion":{"low":5.19,"central":5.19,"high":5.19,"unit":"ratio","basis":"Google Cloud backlog ÷ Google Cloud annualised revenue (segment-wide, includes non-AI contracts)","period":"Backlog at 30 June 2026","periodEnd":"2026-06-30","source":{"url":"https://blog.google/company-news/inside-google/message-ceo/alphabet-earnings-q2-2026/","title":"Q2 2026 earnings call: Remarks from our CEO","publisher":"Google","publishedDate":"2026-07-22","accessedDate":"2026-10-22","tier":"T2","locator":"'And Cloud backlog grew to $514 billion.' (Sundar Pichai, Q2 2026 call remarks, 22 July 2026)"},"note":"The company says Google Cloud backlog grew to $514B; segment revenue is $24,768M in Q2 2026 (Alphabet Form 8-K exhibit 99.1, 2026-07-22), x 4 = $99.1B; 514 / 99.07 = 5.19. Backlog and revenue are both segment-wide, so non-AI contracts (Workspace, security) are inside both numbers; the company attributes the growth to AI infrastructure and AI solutions. Tier is the lower of the two (both T2)."},"baselineRevenueUsd":{"low":99072000000,"central":99072000000,"high":99072000000,"unit":"USD","basis":"Google Cloud segment revenue, latest quarter x 4, US dollars","period":"Q2 2026 annualised","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0001652044/000165204426000066/googexhibit991q22026.htm","title":"Alphabet Announces Second Quarter 2026 Results (Form 8-K exhibit 99.1)","publisher":"Alphabet Inc.","publishedDate":"2026-07-22","accessedDate":"2026-10-22","tier":"T2","locator":"Segment table: Google Cloud revenue Q2 2026 $24,768M"},"note":"Segment revenue of the unit (never Alphabet's total). Only the latest quarter was read, so this is an annualised figure rather than a fiscal-year total."}},"oracle-cloud-infrastructure":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"oracle-cloud-infrastructure","name":"Oracle Cloud Infrastructure (OCI)","kind":"unit","parentId":"oracle","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":23.215,"central":23.215,"high":23.215,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"Q1 FY2027 (June-August 2026), annualised","periodEnd":"2026-08-31","source":{"url":"https://nextplatform.com/cloud/2026/09/14/despite-hefty-capital-investments-the-cloud-starts-paying-off-for-oracle/5296362","title":"Despite Hefty Capital Investments, The Cloud Starts Paying Off For Oracle","publisher":"The Next Platform","publishedDate":"2026-09-14","accessedDate":"2026-10-22","tier":"T5","locator":"'Sales of CPU and GPU capacity comprise around 88 percent of Oracle's cloud IaaS sales', $6.5B in the quarter, up 2.5X; OCI revenue $7,388M vs $3,347M (Oracle release, 2026-09-10); more than 300,000 GPUs delivered in the quarter"},"note":"5LayerCake T5 estimate. Oracle's CPU and GPU compute revenue was $6.5B in June-August 2026, 88% of IaaS (The Next Platform, T3); x 4 = $26.0B. CPU compute is assumed about $1.5B a quarter ($6.0B a year, a 5LayerCake assumption; for scale, the whole of IaaS was $3.3B a quarter a year earlier), leaving about $20B of GPU rental. Oracle reports no GPU revenue; its release says it delivered more than 300,000 GPUs in the quarter and that GPU utilisation was 97.9% (Next Platform). 20.0 / 86.2. Denominator (the same in every file of this sublayer): GPU computing outside mainland China = about $86.2B a year, annualised from Q2 2026 (Oracle: June-August 2026). It is the sum of the provider estimates: AWS 11.0, Azure 18.0 and Google Cloud 9.0 (assumed splits of company-reported AI figures; none reports GPU rental), Oracle OCI 20.0, CoreWeave 10.3 (Q2 revenue $2,575M x 4), Nebius 2.3 (AI cloud revenue $574.9M x 4), all other registered providers 9.6, and 6.0 for providers outside the register. Reconciliation with published totals: Gartner forecasts AI-optimized IaaS spending of $42.3B for all of 2026 (Techstrong, 2026-09-21; Gartner's 10 August 2026 release is titled as a worldwide forecast; gartner.com returned HTTP 403 and was not retried, so its definition is unread). CoreWeave 10.3 + Nebius 2.3 + Oracle's CPU and GPU compute revenue of 26.0 ($6.5B in June-August 2026 x 4, The Next Platform, 2026-09-14) already total 38.6, 91% of Gartner's whole-year worldwide figure, before AWS, Microsoft, Google or any other provider. Gartner's category must therefore be narrower than AI compute rented to customers (or low), so it is not used as the denominator; it is the lowest case. Synergy's neocloud revenue above $25B for 2025 (2026-04-02) counts GPU-as-a-service, generative-AI platforms and high-density data-centre capacity and lists OpenAI and Anthropic among cloud providers, so it is wider than rented GPUs; the registered neoclouds' rental revenue sums to about 19.4 here. 5LayerCake judgement of the plausible range for the total: about $58B (every provider at the low end of its own range) to about $115B; the T5 share width covers it. Hyperscalers resell some capacity bought from neoclouds, so the total can double count by a few percent."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://investor.oracle.com/investor-news/news-details/2026/Oracle-Announces-Q1-Results-Driven-by-Triple-Digit-Growth-in-Cloud-Infrastructure-Revenues/default.aspx","title":"Oracle Announces Q1 Results Driven by Triple Digit Growth in Cloud Infrastructure Revenues","publisher":"Oracle Corporation","publishedDate":"2026-09-10","accessedDate":"2026-10-22","tier":"T2","locator":"Q1 FY27 release describes AI Cloud capacity as GPUs (300,000 delivered) and no Oracle-designed accelerator"}},"china":{"participates":false,"evidence":{"url":"https://docs.oracle.com/en-us/iaas/Content/General/Concepts/regions.htm","title":"Regions and Availability Domains","publisher":"Oracle","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"OCI commercial region table lists 12 Asia Pacific regions (Australia, India, Indonesia, Japan, Malaysia, Singapore, South Korea) and no mainland China region"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"6791af5cb89e824529bc8694cc3b39632c368ea7fc6a265638314a9e055a8be2"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"6791af5cb89e824529bc8694cc3b39632c368ea7fc6a265638314a9e055a8be2"}},"priorShares":{"gpu-capacity":{"low":18.2,"central":18.2,"high":18.2,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"Q1 FY2026 (June-August 2025), annualised","periodEnd":"2025-08-31","source":{"url":"https://nextplatform.com/cloud/2026/09/14/despite-hefty-capital-investments-the-cloud-starts-paying-off-for-oracle/5296362","title":"Despite Hefty Capital Investments, The Cloud Starts Paying Off For Oracle","publisher":"The Next Platform","publishedDate":"2026-09-14","accessedDate":"2026-10-22","tier":"T5","locator":"Compute (CPU and GPU) revenue $6.5B in June-August 2026, 'up 2.5X year on year' (prior $2.6B); Oracle release: Cloud Infra (IaaS) $3,347M in Q1 FY26"},"note":"5LayerCake T5 derivation. Prior-year compute revenue $2.6B a quarter (6.5 / 2.5) x 4 = $10.4B; the same 77% GPU fraction as now (20.0 / 26.0) gives GPU rental of about $8.0B. A year earlier the GPU fraction was probably lower (more of compute was CPU), so this prior is high and the share gain is understated. Prior-period total (about 44.0B) = 86.2 / 1.96, using Gartner's forecast that AI-optimized IaaS spending rises 96% in 2026 over 2025 (T4; Techstrong, 2026-09-21). Synergy's 165% year-on-year growth in GenAI-specific cloud services in Q2 2026 (2026-07-30) is the higher-growth alternative (prior total about 32.5B), which would raise every prior share and shrink every share gain; the same prior total is used for every company so the ranking of share changes is little affected. 8.0 / 44.0."}},"activityGrowth":{"current":{"low":6500000000,"central":6500000000,"high":6500000000,"unit":"USD","basis":"Oracle cloud infrastructure compute (CPU and GPU) revenue for the June-August quarter, US dollars","period":"Q1 FY2027 (June-August 2026)","periodEnd":"2026-08-31","source":{"url":"https://nextplatform.com/cloud/2026/09/14/despite-hefty-capital-investments-the-cloud-starts-paying-off-for-oracle/5296362","title":"Despite Hefty Capital Investments, The Cloud Starts Paying Off For Oracle","publisher":"The Next Platform","publishedDate":"2026-09-14","accessedDate":"2026-10-22","tier":"T3","locator":"'Sales of CPU and GPU capacity comprise around 88 percent of Oracle's cloud IaaS sales' generating $6.5 billion, up 2.5X year on year"},"note":"The Next Platform's figure (T3), 88% of the $7,388M IaaS revenue in Oracle's release. Compute is mostly AI GPU capacity but includes CPU capacity."},"prior":{"low":2600000000,"central":2600000000,"high":2600000000,"unit":"USD","basis":"Oracle cloud infrastructure compute (CPU and GPU) revenue for the June-August quarter, US dollars","period":"Q1 FY2026 (June-August 2025)","periodEnd":"2025-08-31","source":{"url":"https://nextplatform.com/cloud/2026/09/14/despite-hefty-capital-investments-the-cloud-starts-paying-off-for-oracle/5296362","title":"Despite Hefty Capital Investments, The Cloud Starts Paying Off For Oracle","publisher":"The Next Platform","publishedDate":"2026-09-14","accessedDate":"2026-10-22","tier":"T3","locator":"Compute revenue 'up 2.5X year on year'"},"note":"Derived: 6.5 / 2.5 = $2.6B. Cross-check from the company: IaaS revenue grew 121% ($3,347M to $7,388M)."}},"committedExpansion":{"low":22.47,"central":22.47,"high":22.47,"unit":"ratio","basis":"Oracle-wide remaining performance obligations (all products, including software and support; not OCI-specific) ÷ OCI annualised IaaS revenue (stand-in)","period":"RPO at 31 August 2026","periodEnd":"2026-08-31","source":{"url":"https://investor.oracle.com/investor-news/news-details/2026/Oracle-Announces-Q1-Results-Driven-by-Triple-Digit-Growth-in-Cloud-Infrastructure-Revenues/default.aspx","title":"Oracle Announces Q1 Results Driven by Triple Digit Growth in Cloud Infrastructure Revenues","publisher":"Oracle Corporation","publishedDate":"2026-09-10","accessedDate":"2026-10-22","tier":"T2","locator":"'Oracle booked more than $30 billion of additional AI cloud contracts in Q1 increasing its RPO to $664 billion' (RPO is Oracle-wide; the release gives no OCI-specific backlog); Cloud Infrastructure (IaaS) $7,388M"},"note":"Stand-in, labelled off-basis: Oracle publishes no OCI-specific backlog. Remaining performance obligations were $664B at 31 August 2026, up $209B on a year earlier (Oracle release, 2026-09-10, T2); the Form 10-Q (https://www.sec.gov/Archives/edgar/data/0001341439/000119312526389274/orcl-20260831.htm) gives the same total and only the timing (about 13% within twelve months, 37% in months 13 to 36, 34% in months 37 to 60), with no split by product. $664B ÷ annualised IaaS revenue (7.388 x 4 = $29.55B) = 22.47. RPO also covers SaaS and license support, so the ratio overstates OCI's own by an unknown amount; the engine widens an off-basis observation by 25 points. The company says it booked more than $30B of additional AI cloud contracts in Q1, and The Next Platform reports the CFO saying about half of RPO converts within 36 months, consistent with the 10-Q's 13% plus 37%. Both inputs are T2 company statements, so the tier is T2 (the lower of the two).","offBasis":true},"baselineRevenueUsd":{"low":18101000000,"central":18101000000,"high":18101000000,"unit":"USD","basis":"Oracle Cloud Infrastructure (IaaS) revenue, fiscal year, US dollars","period":"FY2026 (June 2025 - May 2026)","periodEnd":"2026-05-31","source":{"url":"https://investor.oracle.com/investor-news/news-details/2026/Oracle-Announces-Q1-Results-Driven-by-Triple-Digit-Growth-in-Cloud-Infrastructure-Revenues/default.aspx","title":"Oracle Announces Q1 Results Driven by Triple Digit Growth in Cloud Infrastructure Revenues","publisher":"Oracle Corporation","publishedDate":"2026-09-10","accessedDate":"2026-10-22","tier":"T2","locator":"Supplemental revenue table: Cloud Infrastructure quarterly series $3,347M, $4,079M, $4,888M, $5,787M (sum $18,101M)"},"note":"Segment line of the unit (never Oracle's total revenue); sum of the four fiscal-2026 quarters shown in the release's table."}},"microsoft-intelligent-cloud":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"microsoft-intelligent-cloud","name":"Microsoft Azure","kind":"unit","parentId":"microsoft","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":20.893,"central":20.893,"high":20.893,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"Q2 2026 annualised (AI run rate as of March 2026, rolled forward)","periodEnd":"2026-06-30","source":{"url":"https://www.microsoft.com/en-us/investor/earnings/fy-2026-q3/press-release-webcast","title":"Microsoft FY26 Q3 earnings release (April 29, 2026)","publisher":"Microsoft Corporation","publishedDate":"2026-04-29","accessedDate":"2026-10-22","tier":"T5","locator":"Release quote: 'Our AI business surpassed an annual revenue run rate of $37 billion, up 123% year-over-year'; Q4 FY26 call: Azure surpassed $100B in FY26, up 41%"},"note":"5LayerCake T5 estimate (assumed split). The company says its AI business (Copilots, Foundry model APIs and Azure AI infrastructure) passed a $37B annual run rate in March 2026, up 123% (release, 2026-04-29), and Azure surpassed $100B in FY2026 (call, 2026-07-29); it gives no split. Assumed $41B by June, of which GPU infrastructure rented to AI labs and customers about 44% = $18B; Copilot subscriptions and per-token model APIs are excluded. Microsoft says it added a gigawatt of capacity in the quarter and that demand exceeds supply. 18.0 / 86.2. Denominator (the same in every file of this sublayer): GPU computing outside mainland China = about $86.2B a year, annualised from Q2 2026 (Oracle: June-August 2026). It is the sum of the provider estimates: AWS 11.0, Azure 18.0 and Google Cloud 9.0 (assumed splits of company-reported AI figures; none reports GPU rental), Oracle OCI 20.0, CoreWeave 10.3 (Q2 revenue $2,575M x 4), Nebius 2.3 (AI cloud revenue $574.9M x 4), all other registered providers 9.6, and 6.0 for providers outside the register. Reconciliation with published totals: Gartner forecasts AI-optimized IaaS spending of $42.3B for all of 2026 (Techstrong, 2026-09-21; Gartner's 10 August 2026 release is titled as a worldwide forecast; gartner.com returned HTTP 403 and was not retried, so its definition is unread). CoreWeave 10.3 + Nebius 2.3 + Oracle's CPU and GPU compute revenue of 26.0 ($6.5B in June-August 2026 x 4, The Next Platform, 2026-09-14) already total 38.6, 91% of Gartner's whole-year worldwide figure, before AWS, Microsoft, Google or any other provider. Gartner's category must therefore be narrower than AI compute rented to customers (or low), so it is not used as the denominator; it is the lowest case. Synergy's neocloud revenue above $25B for 2025 (2026-04-02) counts GPU-as-a-service, generative-AI platforms and high-density data-centre capacity and lists OpenAI and Anthropic among cloud providers, so it is wider than rented GPUs; the registered neoclouds' rental revenue sums to about 19.4 here. 5LayerCake judgement of the plausible range for the total: about $58B (every provider at the low end of its own range) to about $115B; the T5 share width covers it. Hyperscalers resell some capacity bought from neoclouds, so the total can double count by a few percent."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://blogs.microsoft.com/blog/2026/01/26/maia-200-the-ai-accelerator-built-for-inference/","title":"Maia 200: The AI accelerator built for inference","publisher":"Microsoft","publishedDate":"2026-01-26","accessedDate":"2026-10-22","tier":"T2","locator":"Maia 200 is described as serving Microsoft's own models and Superintelligence team; the article offers a Maia SDK preview and does not offer Maia as rentable instances"}},"china":{"participates":false,"evidence":{"url":"https://learn.microsoft.com/en-us/azure/china/overview-operations","title":"Microsoft Azure in China","publisher":"Microsoft Learn","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Azure in China is 'independently operated and transacted by' 21Vianet, so Microsoft does not rent AI compute in mainland China itself"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"8e30e27d32a78a5abd0fd1c40ee831f828ac35dde334511455f141fcdf5aca76"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"8e30e27d32a78a5abd0fd1c40ee831f828ac35dde334511455f141fcdf5aca76"}},"priorShares":{"gpu-capacity":{"low":16.572,"central":16.572,"high":16.572,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"March 2025 AI run rate, GPU part","periodEnd":"2025-03-31","source":{"url":"https://www.microsoft.com/en-us/investor/earnings/fy-2026-q3/press-release-webcast","title":"Microsoft FY26 Q3 earnings release (April 29, 2026)","publisher":"Microsoft Corporation","publishedDate":"2026-04-29","accessedDate":"2026-10-22","tier":"T5","locator":"'Our AI business surpassed an annual revenue run rate of $37 billion, up 123% year-over-year.' (Satya Nadella, 29 April 2026)"},"note":"5LayerCake T5 derivation. AI business run rate in March 2025 = 37 / 2.23 = $16.6B; the same 44% GPU fraction as now (18.0 / 41.0) gives about $7.3B. The fraction is an assumption and may have been higher then (Copilot was smaller), so the prior is high and the share gain understated. Prior-period total (about 44.0B) = 86.2 / 1.96, using Gartner's forecast that AI-optimized IaaS spending rises 96% in 2026 over 2025 (T4; Techstrong, 2026-09-21). Synergy's 165% year-on-year growth in GenAI-specific cloud services in Q2 2026 (2026-07-30) is the higher-growth alternative (prior total about 32.5B), which would raise every prior share and shrink every share gain; the same prior total is used for every company so the ranking of share changes is little affected. 7.3 / 44.0."}},"activityGrowth":{"current":{"low":37000000000,"central":37000000000,"high":37000000000,"unit":"USD","basis":"Microsoft AI business annual revenue run rate, US dollars (includes Copilots, Foundry model APIs and Azure AI infrastructure)","period":"March 2026","periodEnd":"2026-03-31","source":{"url":"https://www.microsoft.com/en-us/investor/earnings/fy-2026-q3/press-release-webcast","title":"Microsoft FY26 Q3 earnings release (April 29, 2026)","publisher":"Microsoft Corporation","publishedDate":"2026-04-29","accessedDate":"2026-10-22","tier":"T2","locator":"'Our AI business surpassed an annual revenue run rate of $37 billion, up 123% year-over-year.'"},"note":"The company says; a floor ('surpassed'). Wider than rented compute because it includes Copilot subscriptions and model APIs; no narrower figure is published."},"prior":{"low":16592000000,"central":16592000000,"high":16592000000,"unit":"USD","basis":"Microsoft AI business annual revenue run rate, US dollars (includes Copilots, Foundry model APIs and Azure AI infrastructure)","period":"March 2025","periodEnd":"2025-03-31","source":{"url":"https://www.microsoft.com/en-us/investor/earnings/fy-2026-q3/press-release-webcast","title":"Microsoft FY26 Q3 earnings release (April 29, 2026)","publisher":"Microsoft Corporation","publishedDate":"2026-04-29","accessedDate":"2026-10-22","tier":"T2","locator":"Growth of 123% year over year"},"note":"Derived from the company's growth rate: 37 / (1 + 1.23) = $16.6B."}},"baselineRevenueUsd":{"low":100000000000,"central":100000000000,"high":100000000000,"unit":"USD","basis":"Azure and other cloud services revenue, fiscal year, US dollars","period":"FY2026 (July 2025 - June 2026)","periodEnd":"2026-06-30","source":{"url":"https://www.microsoft.com/en-us/investor/events/fy-2026/earnings-fy-2026-q4","title":"FY2026 Q4 earnings conference call","publisher":"Microsoft Corporation","publishedDate":"2026-07-29","accessedDate":"2026-10-22","tier":"T2","locator":"'And Azure surpassed $100 billion, up 41%.' (Satya Nadella, FY26 Q4 call)"},"note":"The company says Azure surpassed $100B in FY2026 (a floor); the unit's own line, not Microsoft's total. No committed-expansion input: Microsoft's commercial RPO ($678B at June 2026; the company does not disclose OpenAI's share) covers all Microsoft products, not this unit."}},"coreweave":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"coreweave","name":"CoreWeave","kind":"public","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":11.956,"central":11.956,"high":11.956,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"Q2 2026 annualised","periodEnd":"2026-06-30","source":{"url":"https://investors.coreweave.com/news/news-details/2026/CoreWeave-Reports-Strong-Second-Quarter-2026-Results/","title":"CoreWeave Reports Strong Second Quarter 2026 Results","publisher":"CoreWeave, Inc.","publishedDate":"2026-08-11","accessedDate":"2026-10-22","tier":"T5","locator":"Q2 2026 revenue $2,575M (Q2 2025 $1,212M); backlog about $104B; active power 1.5 GW"},"note":"Company filing: Q2 2026 revenue $2,575M (release, 2026-08-11) x 4 = $10.3B; CoreWeave sells GPU cloud almost entirely, so revenue is taken as rental revenue, all outside mainland China. 10.3 / 86.2. The company's own number is T2; the share is T5 because the denominator is a 5LayerCake estimate. Denominator (the same in every file of this sublayer): GPU computing outside mainland China = about $86.2B a year, annualised from Q2 2026 (Oracle: June-August 2026). It is the sum of the provider estimates: AWS 11.0, Azure 18.0 and Google Cloud 9.0 (assumed splits of company-reported AI figures; none reports GPU rental), Oracle OCI 20.0, CoreWeave 10.3 (Q2 revenue $2,575M x 4), Nebius 2.3 (AI cloud revenue $574.9M x 4), all other registered providers 9.6, and 6.0 for providers outside the register. Reconciliation with published totals: Gartner forecasts AI-optimized IaaS spending of $42.3B for all of 2026 (Techstrong, 2026-09-21; Gartner's 10 August 2026 release is titled as a worldwide forecast; gartner.com returned HTTP 403 and was not retried, so its definition is unread). CoreWeave 10.3 + Nebius 2.3 + Oracle's CPU and GPU compute revenue of 26.0 ($6.5B in June-August 2026 x 4, The Next Platform, 2026-09-14) already total 38.6, 91% of Gartner's whole-year worldwide figure, before AWS, Microsoft, Google or any other provider. Gartner's category must therefore be narrower than AI compute rented to customers (or low), so it is not used as the denominator; it is the lowest case. Synergy's neocloud revenue above $25B for 2025 (2026-04-02) counts GPU-as-a-service, generative-AI platforms and high-density data-centre capacity and lists OpenAI and Anthropic among cloud providers, so it is wider than rented GPUs; the registered neoclouds' rental revenue sums to about 19.4 here. 5LayerCake judgement of the plausible range for the total: about $58B (every provider at the low end of its own range) to about $115B; the T5 share width covers it. Hyperscalers resell some capacity bought from neoclouds, so the total can double count by a few percent."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://www.coreweave.com/","title":"CoreWeave home page","publisher":"CoreWeave, Inc.","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Home page lists NVIDIA hardware (GB200 NVL72, Vera Rubin NVL72); no CoreWeave-designed accelerator"}},"china":{"participates":false,"evidence":{"url":"https://www.coreweave.com/","title":"CoreWeave home page","publisher":"CoreWeave, Inc.","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Page names headquarters in Livingston NJ and an India expansion; it does not mention China"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"2e7b02f649b55f4a706ce866333896314116327ed13df6dd79c80bac252b309f"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"2e7b02f649b55f4a706ce866333896314116327ed13df6dd79c80bac252b309f"}},"priorShares":{"gpu-capacity":{"low":11.029,"central":11.029,"high":11.029,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"Q2 2025 annualised","periodEnd":"2025-06-30","source":{"url":"https://investors.coreweave.com/news/news-details/2026/CoreWeave-Reports-Strong-Second-Quarter-2026-Results/","title":"CoreWeave Reports Strong Second Quarter 2026 Results","publisher":"CoreWeave, Inc.","publishedDate":"2026-08-11","accessedDate":"2026-10-22","tier":"T5","locator":"Q2 2025 revenue $1,212M (comparison column of the Q2 2026 release)"},"note":"5LayerCake T5 derivation: Q2 2025 revenue $1,212M x 4 = $4.85B. Prior-period total (about 44.0B) = 86.2 / 1.96, using Gartner's forecast that AI-optimized IaaS spending rises 96% in 2026 over 2025 (T4; Techstrong, 2026-09-21). Synergy's 165% year-on-year growth in GenAI-specific cloud services in Q2 2026 (2026-07-30) is the higher-growth alternative (prior total about 32.5B), which would raise every prior share and shrink every share gain; the same prior total is used for every company so the ranking of share changes is little affected. 4.85 / 44.0."}},"activityGrowth":{"current":{"low":2575000000,"central":2575000000,"high":2575000000,"unit":"USD","basis":"CoreWeave quarterly revenue, US dollars","period":"Q2 2026","periodEnd":"2026-06-30","source":{"url":"https://investors.coreweave.com/news/news-details/2026/CoreWeave-Reports-Strong-Second-Quarter-2026-Results/","title":"CoreWeave Reports Strong Second Quarter 2026 Results","publisher":"CoreWeave, Inc.","publishedDate":"2026-08-11","accessedDate":"2026-10-22","tier":"T2","locator":"Q2 2026 revenue $2,575M"},"note":"Company results release."},"prior":{"low":1212000000,"central":1212000000,"high":1212000000,"unit":"USD","basis":"CoreWeave quarterly revenue, US dollars","period":"Q2 2025","periodEnd":"2025-06-30","source":{"url":"https://investors.coreweave.com/news/news-details/2026/CoreWeave-Reports-Strong-Second-Quarter-2026-Results/","title":"CoreWeave Reports Strong Second Quarter 2026 Results","publisher":"CoreWeave, Inc.","publishedDate":"2026-08-11","accessedDate":"2026-10-22","tier":"T2","locator":"Q2 2025 revenue $1,212M"},"note":"Company results release, comparison period."}},"committedExpansion":{"low":10.1,"central":10.1,"high":10.1,"unit":"ratio","basis":"CoreWeave revenue backlog ÷ annualised quarterly revenue","period":"Backlog at 30 June 2026","periodEnd":"2026-06-30","source":{"url":"https://investors.coreweave.com/news/news-details/2026/CoreWeave-Reports-Strong-Second-Quarter-2026-Results/","title":"CoreWeave Reports Strong Second Quarter 2026 Results","publisher":"CoreWeave, Inc.","publishedDate":"2026-08-11","accessedDate":"2026-10-22","tier":"T2","locator":"Revenue backlog 'approximately $104 billion as of June 30, 2026'; footnote: excludes more than $25 billion of net new customer commitments added in early Q3"},"note":"The company says backlog was about $104B (remaining performance obligations plus other committed amounts) at 30 June 2026; Q2 revenue $2,575M x 4 = $10.3B; 104 / 10.3 = 10.1. The company adds more than $25B of net new commitments in early Q3, not counted."},"baselineRevenueUsd":{"low":10300000000,"central":10300000000,"high":10300000000,"unit":"USD","basis":"CoreWeave revenue, latest quarter x 4, US dollars","period":"Q2 2026 annualised","periodEnd":"2026-06-30","source":{"url":"https://investors.coreweave.com/news/news-details/2026/CoreWeave-Reports-Strong-Second-Quarter-2026-Results/","title":"CoreWeave Reports Strong Second Quarter 2026 Results","publisher":"CoreWeave, Inc.","publishedDate":"2026-08-11","accessedDate":"2026-10-22","tier":"T2","locator":"Q2 2026 revenue $2,575M"},"note":"Annualised from the latest quarter; only the Q2 release was read."}},"alibaba-cloud":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"alibaba-cloud","name":"Alibaba Cloud","kind":"unit","parentId":"alibaba","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":0.464,"central":0.464,"high":0.464,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"2026; revenue estimated","periodEnd":"2026-09-23","source":{"url":"https://www.alibabacloud.com/press-room/alibaba-cloud-expands-global-infrastructure-and-ai","title":"Alibaba Cloud Expands Global Infrastructure and AI Portfolio to Accelerate Enterprise AI Adoption","publisher":"Alibaba Cloud","publishedDate":"2026-09-23","accessedDate":"2026-10-22","tier":"T5","locator":"'currently operates 107 availability zones in 31 regions around the world'; new regions planned in Turkiye, Finland, Netherlands; no overseas GPU figure"},"note":"5LayerCake T5 estimate. Alibaba's June-quarter 'AI Cloud and Compute Services' revenue was RMB48,437M ($7.1B) but overseas and GPU-rental parts are not disclosed. Assumed overseas AI compute rental about $0.4B. 400 / market total. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://www.alibabacloud.com/press-room/alibaba-cloud-expands-global-infrastructure-and-ai","title":"Alibaba Cloud Expands Global Infrastructure and AI Portfolio to Accelerate Enterprise AI Adoption","publisher":"Alibaba Cloud","publishedDate":"2026-09-23","accessedDate":"2026-10-22","tier":"T2","locator":"Press release describes regions and AI portfolio; own-chip (T-Head) rental outside China is not stated"}},"china":{"low":38.1,"central":38.1,"high":38.1,"unit":"percent","basis":"Share of AI-cloud market revenue in mainland China, calendar year 2025 (Omdia; includes model-as-a-service and platform revenue, wider than rented computing)","offBasis":true,"period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://eu.36kr.com/en/p/3924679520647302","title":"B-End Market Saturation Reached: Untapped C-End Frontier Becomes New Competitive Battlefield for Cloud Vendors to Seize Market Territory (citing Omdia, China AI cloud market 2025)","publisher":"36Kr","publishedDate":"2026-08-04","accessedDate":"2026-10-22","tier":"T4","locator":"Omdia China AI cloud market 2025: Alibaba Cloud 38.1% of RMB56.7B"},"note":"T4 via 36Kr. Omdia's China AI cloud market for 2025 was RMB56.7B with Alibaba Cloud 38.1%, Volcano Engine 20.4%, Baidu Cloud 9.4%, Tencent Cloud 6.3% and eSurfing Cloud (China Telecom) 5.9%; the top five hold more than 80% (36Kr, 2026-08-04, citing Omdia data released 19 May; T4). Omdia's first-half 2025 report (SCMP via Yahoo Finance, 2025-09-10) defines the market as 'computing infrastructure optimised for generative AI tasks' and gave Alibaba 35.8%, Volcano Engine 14.8%, Huawei Cloud 13.1%, Tencent Cloud 7.0% and Baidu Cloud 6.1%. The full-year shares differ in structure (Volcano 14.8 to 20.4, Baidu 6.1 to 9.4, Huawei 13.1 to outside the top five), which suggests the full-year measure also counts model services, so every full-year China share is flagged offBasis; the first-half shares of the four providers named in both reports lie inside the widened ranges. Omdia gives no earlier full-year vendor shares, so there is no prior share on the same basis. Alibaba's own results (June quarter 2026): AI Cloud and Compute Services revenue RMB48,437M (+45%) and AI-related product revenue RMB12,376M with 'more than 100%' growth, the twelfth consecutive quarter of triple-digit growth; the exact growth is not stated, so no activity-growth input is possible."}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2},"china":{"levelLow":2,"levelHigh":4}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"e5919e395cd3d706fc9b49f49bb324eab07a05cd48c187cae2fd0271018aa3bb"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"e5919e395cd3d706fc9b49f49bb324eab07a05cd48c187cae2fd0271018aa3bb"}},"baselineRevenueUsd":{"low":28555300000,"central":28555300000,"high":28555300000,"unit":"USD","basis":"Alibaba AI Cloud and Compute Services segment revenue, latest quarter x 4, US dollars","period":"June quarter 2026 annualised","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/1577552/000110465926099220/tm2623667d1_ex99-1.htm","title":"Alibaba Group Announces June Quarter 2026 Results (Form 6-K exhibit 99.1)","publisher":"Alibaba Group Holding Limited","publishedDate":"2026-08-20","accessedDate":"2026-10-22","tier":"T2","locator":"AI Cloud and Compute Services revenue RMB48,437 million (US$7,139 million) in the June 2026 quarter"},"note":"RMB48,437M x 4 = RMB193.7B converted at the release's own rate (US$7,139M for RMB48,437M = 6.785). The segment now combines Cloud Intelligence Group and T-Head, so it is wider than rented compute. Segment of the unit, not Alibaba's total."}},"volcano-engine":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"volcano-engine","name":"Volcano Engine (ByteDance)","kind":"unit","parentId":"bytedance","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":0.116,"central":0.116,"high":0.116,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"2026; revenue estimated","periodEnd":"2026-10-22","source":{"url":"https://www.byteplus.com/","title":"BytePlus site (GPU Compute Service listing)","publisher":"BytePlus (ByteDance)","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"BytePlus lists a 'GPU Compute Service'; console links point to ap-southeast-1; no scale"},"note":"5LayerCake T5 estimate for BytePlus, Volcano Engine's international arm; no revenue or fleet disclosed. Assumed $0.1B. 100 / market total. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://www.byteplus.com/","title":"BytePlus site (GPU Compute Service listing)","publisher":"BytePlus (ByteDance)","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Page lists a GPU Compute Service; no ByteDance-designed accelerator for rental abroad"}},"china":{"low":20.4,"central":20.4,"high":20.4,"unit":"percent","basis":"Share of AI-cloud market revenue in mainland China, calendar year 2025 (Omdia; includes model-as-a-service and platform revenue, wider than rented computing)","offBasis":true,"period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://eu.36kr.com/en/p/3924679520647302","title":"B-End Market Saturation Reached: Untapped C-End Frontier Becomes New Competitive Battlefield for Cloud Vendors to Seize Market Territory (citing Omdia, China AI cloud market 2025)","publisher":"36Kr","publishedDate":"2026-08-04","accessedDate":"2026-10-22","tier":"T4","locator":"Omdia China AI cloud market 2025: Volcano Engine 20.4%"},"note":"T4 via 36Kr. Omdia's China AI cloud market for 2025 was RMB56.7B with Alibaba Cloud 38.1%, Volcano Engine 20.4%, Baidu Cloud 9.4%, Tencent Cloud 6.3% and eSurfing Cloud (China Telecom) 5.9%; the top five hold more than 80% (36Kr, 2026-08-04, citing Omdia data released 19 May; T4). Omdia's first-half 2025 report (SCMP via Yahoo Finance, 2025-09-10) defines the market as 'computing infrastructure optimised for generative AI tasks' and gave Alibaba 35.8%, Volcano Engine 14.8%, Huawei Cloud 13.1%, Tencent Cloud 7.0% and Baidu Cloud 6.1%. The full-year shares differ in structure (Volcano 14.8 to 20.4, Baidu 6.1 to 9.4, Huawei 13.1 to outside the top five), which suggests the full-year measure also counts model services, so every full-year China share is flagged offBasis; the first-half shares of the four providers named in both reports lie inside the widened ranges. Omdia gives no earlier full-year vendor shares, so there is no prior share on the same basis. Volcano Engine is part of ByteDance, publishes no revenue and has no filings, so no baseline revenue, growth or backlog input exists."}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2},"china":{"levelLow":2,"levelHigh":4}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"8b14487abb731c1791651f4517d64a7230862899be3b08dd2b059fad1282c0fd"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"8b14487abb731c1791651f4517d64a7230862899be3b08dd2b059fad1282c0fd"}}},"nebius":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"nebius","name":"Nebius","kind":"public","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":2.67,"central":2.67,"high":2.67,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"Q2 2026 annualised","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0001513845/000110465926094568/tm2622968d1_ex99-2.htm","title":"Nebius Group second quarter 2026 shareholder letter (Form 6-K exhibit 99.2)","publisher":"Nebius Group N.V.","publishedDate":"2026-08-12","accessedDate":"2026-10-22","tier":"T5","locator":"AI cloud segment revenue Q2 2026 $574.9M; group $582.3M; ARR $3.0B at end of June"},"note":"Company statement: AI cloud revenue $574.9M in Q2 2026 (up 514%) x 4 = $2.30B; the end-June ARR of $3.0B is a run rate on June revenue, is higher and is not used as central. 2.3 / 86.2. Denominator (the same in every file of this sublayer): GPU computing outside mainland China = about $86.2B a year, annualised from Q2 2026 (Oracle: June-August 2026). It is the sum of the provider estimates: AWS 11.0, Azure 18.0 and Google Cloud 9.0 (assumed splits of company-reported AI figures; none reports GPU rental), Oracle OCI 20.0, CoreWeave 10.3 (Q2 revenue $2,575M x 4), Nebius 2.3 (AI cloud revenue $574.9M x 4), all other registered providers 9.6, and 6.0 for providers outside the register. Reconciliation with published totals: Gartner forecasts AI-optimized IaaS spending of $42.3B for all of 2026 (Techstrong, 2026-09-21; Gartner's 10 August 2026 release is titled as a worldwide forecast; gartner.com returned HTTP 403 and was not retried, so its definition is unread). CoreWeave 10.3 + Nebius 2.3 + Oracle's CPU and GPU compute revenue of 26.0 ($6.5B in June-August 2026 x 4, The Next Platform, 2026-09-14) already total 38.6, 91% of Gartner's whole-year worldwide figure, before AWS, Microsoft, Google or any other provider. Gartner's category must therefore be narrower than AI compute rented to customers (or low), so it is not used as the denominator; it is the lowest case. Synergy's neocloud revenue above $25B for 2025 (2026-04-02) counts GPU-as-a-service, generative-AI platforms and high-density data-centre capacity and lists OpenAI and Anthropic among cloud providers, so it is wider than rented GPUs; the registered neoclouds' rental revenue sums to about 19.4 here. 5LayerCake judgement of the plausible range for the total: about $58B (every provider at the low end of its own range) to about $115B; the T5 share width covers it. Hyperscalers resell some capacity bought from neoclouds, so the total can double count by a few percent."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://nebius.com/","title":"Nebius home page","publisher":"Nebius","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Home page: non-virtualized NVIDIA GPUs (HGX B300); no Nebius-designed accelerator"}},"china":{"participates":false,"evidence":{"url":"https://nebius.com/","title":"Nebius home page","publisher":"Nebius","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Page does not mention China; corporate entity Nebius B.V.; programmes for Europe's AI builders"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"0c5d1c41de12548e728db8a6505d0247c78692b8b83f12662e023466a605a8ba"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"0c5d1c41de12548e728db8a6505d0247c78692b8b83f12662e023466a605a8ba"}},"priorShares":{"gpu-capacity":{"low":0.852,"central":0.852,"high":0.852,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"Q2 2025 annualised","periodEnd":"2025-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0001513845/000110465926094568/tm2622968d1_ex99-2.htm","title":"Nebius Group second quarter 2026 shareholder letter (Form 6-K exhibit 99.2)","publisher":"Nebius Group N.V.","publishedDate":"2026-08-12","accessedDate":"2026-10-22","tier":"T5","locator":"'Nebius AI cloud revenue was $574.9 million in Q2'26, a 514% increase year-over-year from Q2'25.'"},"note":"5LayerCake T5 derivation: Q2 2025 AI cloud revenue = 574.9 / (1 + 5.14) = $93.6M, x 4 = $0.37B. Prior-period total (about 44.0B) = 86.2 / 1.96, using Gartner's forecast that AI-optimized IaaS spending rises 96% in 2026 over 2025 (T4; Techstrong, 2026-09-21). Synergy's 165% year-on-year growth in GenAI-specific cloud services in Q2 2026 (2026-07-30) is the higher-growth alternative (prior total about 32.5B), which would raise every prior share and shrink every share gain; the same prior total is used for every company so the ranking of share changes is little affected. 0.37 / 44.0."}},"activityGrowth":{"current":{"low":574900000,"central":574900000,"high":574900000,"unit":"USD","basis":"Nebius AI cloud quarterly revenue, US dollars","period":"Q2 2026","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0001513845/000110465926094568/tm2622968d1_ex99-2.htm","title":"Nebius Group second quarter 2026 shareholder letter (Form 6-K exhibit 99.2)","publisher":"Nebius Group N.V.","publishedDate":"2026-08-12","accessedDate":"2026-10-22","tier":"T2","locator":"Nebius AI cloud revenue $574.9M in Q2'26"},"note":"Company shareholder letter."},"prior":{"low":93631922,"central":93631922,"high":93631922,"unit":"USD","basis":"Nebius AI cloud quarterly revenue, US dollars","period":"Q2 2025","periodEnd":"2025-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0001513845/000110465926094568/tm2622968d1_ex99-2.htm","title":"Nebius Group second quarter 2026 shareholder letter (Form 6-K exhibit 99.2)","publisher":"Nebius Group N.V.","publishedDate":"2026-08-12","accessedDate":"2026-10-22","tier":"T2","locator":"514% increase year-over-year from Q2'25"},"note":"Derived from the company's growth rate: 574.9 / 6.14 = $93.6M."}},"committedExpansion":{"low":17.4,"central":17.4,"high":17.4,"unit":"ratio","basis":"Nebius customer commitments ÷ annualised AI cloud revenue","period":"Commitments at August 2026","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0001513845/000110465926094568/tm2622968d1_ex99-2.htm","title":"Nebius Group second quarter 2026 shareholder letter (Form 6-K exhibit 99.2)","publisher":"Nebius Group N.V.","publishedDate":"2026-08-12","accessedDate":"2026-10-22","tier":"T2","locator":"'We have more than $40 billion of customer commitments.'"},"note":"The company says more than $40B of customer commitments (a floor; the letter names the Microsoft contract, with all capacity tranches delivered, and a second Meta agreement with capacity due in early 2027); AI cloud revenue 574.9 x 4 = $2.30B; 40 / 2.30 = 17.4. The company also expects over $9B of customer prepayments in 2026 and 5 GW of contracted power by the end of 2026."},"baselineRevenueUsd":{"low":2299600000,"central":2299600000,"high":2299600000,"unit":"USD","basis":"Nebius AI cloud revenue, latest quarter x 4, US dollars","period":"Q2 2026 annualised","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0001513845/000110465926094568/tm2622968d1_ex99-2.htm","title":"Nebius Group second quarter 2026 shareholder letter (Form 6-K exhibit 99.2)","publisher":"Nebius Group N.V.","publishedDate":"2026-08-12","accessedDate":"2026-10-22","tier":"T2","locator":"AI cloud revenue $574.9M; group revenue $582.3M"},"note":"The AI cloud line only; annualised from the latest quarter."}},"lambda":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"lambda","name":"Lambda","kind":"private","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":1.741,"central":1.741,"high":1.741,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"CY2026, management projection reported by Bloomberg","periodEnd":"2026-12-31","source":{"url":"https://aiweekly.co/alerts/nvidia-backed-lambda-in-talks-for-3b-pre-ipo-round-at-12b-valuation-as-2026","title":"Nvidia-Backed Lambda in Talks for $3B Pre-IPO Round at $12B+ Valuation as 2026 Revenue Nears $1.5B (reporting Bloomberg)","publisher":"AI Weekly","publishedDate":"2026-08-25","accessedDate":"2026-10-22","tier":"T5","locator":"'Sources say 2026 revenue is on track to top $1.5B' (Bloomberg, as reported by AI Weekly, 25 Aug 2026)"},"note":"Press report of a Bloomberg figure for calendar 2026 (T3 for the revenue, a projection of 'more than $1.5B'; Sacra calls it management guidance); the share is T5 because the denominator is a 5LayerCake estimate. Lambda is private and publishes no revenue. Sacra (https://sacra.com/research/no-side-quests-for-lambda/, 2026-04-20) says Lambda's on-premises GPU workstation and server business, about $100M as of 2024, was shut down in August 2025, so 2026 revenue is GPU cloud rental; the Bloomberg figure is total revenue and is taken as rental on that basis (Sacra's company page also says cloud GPU rental 'contributes the bulk of revenue'). 1,500 / 86,152. Denominator (the same in every file of this sublayer): GPU computing outside mainland China = about $86.2B a year, annualised from Q2 2026 (Oracle: June-August 2026). It is the sum of the provider estimates: AWS 11.0, Azure 18.0 and Google Cloud 9.0 (assumed splits of company-reported AI figures; none reports GPU rental), Oracle OCI 20.0, CoreWeave 10.3 (Q2 revenue $2,575M x 4), Nebius 2.3 (AI cloud revenue $574.9M x 4), all other registered providers 9.6, and 6.0 for providers outside the register. Reconciliation with published totals: Gartner forecasts AI-optimized IaaS spending of $42.3B for all of 2026 (Techstrong, 2026-09-21; Gartner's 10 August 2026 release is titled as a worldwide forecast; gartner.com returned HTTP 403 and was not retried, so its definition is unread). CoreWeave 10.3 + Nebius 2.3 + Oracle's CPU and GPU compute revenue of 26.0 ($6.5B in June-August 2026 x 4, The Next Platform, 2026-09-14) already total 38.6, 91% of Gartner's whole-year worldwide figure, before AWS, Microsoft, Google or any other provider. Gartner's category must therefore be narrower than AI compute rented to customers (or low), so it is not used as the denominator; it is the lowest case. Synergy's neocloud revenue above $25B for 2025 (2026-04-02) counts GPU-as-a-service, generative-AI platforms and high-density data-centre capacity and lists OpenAI and Anthropic among cloud providers, so it is wider than rented GPUs; the registered neoclouds' rental revenue sums to about 19.4 here. 5LayerCake judgement of the plausible range for the total: about $58B (every provider at the low end of its own range) to about $115B; the T5 share width covers it. Hyperscalers resell some capacity bought from neoclouds, so the total can double count by a few percent."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://lambda.ai/","title":"Lambda home page","publisher":"Lambda","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Home page lists NVIDIA GB300 NVL72 superclusters; no Lambda-designed accelerator"}},"china":{"participates":false,"evidence":{"url":"https://lambda.ai/","title":"Lambda home page","publisher":"Lambda","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Page does not mention China or any China site"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"d4f39cb76391002d233250cde72b9bae715e67a9bda7dfd07719eb5a2f97ebfd"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"d4f39cb76391002d233250cde72b9bae715e67a9bda7dfd07719eb5a2f97ebfd"}},"priorShares":{"gpu-capacity":{"low":1.593,"central":1.593,"high":1.593,"unit":"percent","basis":"Lambda total revenue including on-premises hardware sales until August 2025 (rental alone is not separable), calendar year, US dollars, as a share of the rented-GPU denominator","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://sacra.com/c/lambda-labs/","title":"Lambda Labs revenue estimates (company research page)","publisher":"Sacra","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Sacra: 'generated approximately $700M in calendar year 2025' (about $760M annualised by December 2025; about $425M in 2024)"},"note":"5LayerCake T5 derivation from a Sacra estimate (T4) of calendar-2025 revenue, total revenue including on-premises hardware sales, which are not rental (about $100M in 2024 and shut down in August 2025, per Sacra, 2026-04-20: https://sacra.com/research/no-side-quests-for-lambda/); rental alone is not separable, so this prior is flagged offBasis. Prior-period total (about 44.0B) = 86.2 / 1.96, using Gartner's forecast that AI-optimized IaaS spending rises 96% in 2026 over 2025 (T4; Techstrong, 2026-09-21). Synergy's 165% year-on-year growth in GenAI-specific cloud services in Q2 2026 (2026-07-30) is the higher-growth alternative (prior total about 32.5B), which would raise every prior share and shrink every share gain; the same prior total is used for every company so the ranking of share changes is little affected. 0.70 / 44.0.","offBasis":true}},"activityGrowth":{"current":{"low":1500000000,"central":1500000000,"high":1500000000,"unit":"USD","basis":"Lambda total revenue (GPU cloud rental; periods before September 2025 also include on-premises hardware sales), calendar year, US dollars","period":"CY2026 projection","periodEnd":"2026-12-31","source":{"url":"https://aiweekly.co/alerts/nvidia-backed-lambda-in-talks-for-3b-pre-ipo-round-at-12b-valuation-as-2026","title":"Nvidia-Backed Lambda in Talks for $3B Pre-IPO Round at $12B+ Valuation as 2026 Revenue Nears $1.5B (reporting Bloomberg)","publisher":"AI Weekly","publishedDate":"2026-08-25","accessedDate":"2026-10-22","tier":"T3","locator":"2026 revenue on track to top $1.5B (Bloomberg)"},"note":"A projection reported from unnamed sources, not an audited figure; 'more than $1.5B', so a floor."},"prior":{"low":700000000,"central":700000000,"high":700000000,"unit":"USD","basis":"Lambda total revenue (GPU cloud rental; periods before September 2025 also include on-premises hardware sales), calendar year, US dollars","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://sacra.com/c/lambda-labs/","title":"Lambda Labs revenue estimates (company research page)","publisher":"Sacra","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T4","locator":"Approximately $700M in calendar year 2025"},"note":"Sacra estimate (T4); Lambda publishes no figure. It includes on-premises hardware sales until August 2025 (about $100M in 2024, per Sacra), so a rental-only prior would be lower and the true growth in rental revenue somewhat larger than the 2.1x shown; the offBasis flag widens the range.","offBasis":true}},"baselineRevenueUsd":{"low":700000000,"central":700000000,"high":700000000,"unit":"USD","basis":"Lambda total revenue (GPU cloud rental plus on-premises hardware sales until August 2025), calendar year, US dollars","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://sacra.com/c/lambda-labs/","title":"Lambda Labs revenue estimates (company research page)","publisher":"Sacra","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T4","locator":"Approximately $700M in calendar year 2025"},"note":"Sacra estimate; includes hardware sales until August 2025 as well as cloud. Anthropic's reported $35B agreement (WSJ via Reuters, 2026-09-01; term undisclosed) cannot be turned into an annual ratio, so there is no committed-expansion input."}},"baidu-ai-cloud":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"baidu-ai-cloud","name":"Baidu AI Cloud","kind":"unit","parentId":"baidu","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":0.023,"central":0.023,"high":0.023,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"Q2 2026; overseas estimated","periodEnd":"2026-06-30","source":{"url":"https://www.prnewswire.com/news-releases/baidu-announces-second-quarter-2026-results-302853860.html","title":"Baidu Announces Second Quarter 2026 Results","publisher":"Baidu, Inc.","publishedDate":"2026-08-18","accessedDate":"2026-10-22","tier":"T5","locator":"AI Cloud Infra revenue RMB7.3B in Q2 2026; GPU Cloud a component, revenue +283%"},"note":"5LayerCake T5 estimate: Baidu's AI Cloud is China-focused and no overseas GPU rental is reported. Assumed negligible, $0.02B. 20 / market total. Unverified. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://www.prnewswire.com/news-releases/baidu-announces-second-quarter-2026-results-302853860.html","title":"Baidu Announces Second Quarter 2026 Results","publisher":"Baidu, Inc.","publishedDate":"2026-08-18","accessedDate":"2026-10-22","tier":"T2","locator":"Release does not mention Kunlun chips or own-chip rental abroad"}},"china":{"low":9.4,"central":9.4,"high":9.4,"unit":"percent","basis":"Share of AI-cloud market revenue in mainland China, calendar year 2025 (Omdia; includes model-as-a-service and platform revenue, wider than rented computing)","offBasis":true,"period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://eu.36kr.com/en/p/3924679520647302","title":"B-End Market Saturation Reached: Untapped C-End Frontier Becomes New Competitive Battlefield for Cloud Vendors to Seize Market Territory (citing Omdia, China AI cloud market 2025)","publisher":"36Kr","publishedDate":"2026-08-04","accessedDate":"2026-10-22","tier":"T4","locator":"Omdia China AI cloud market 2025: Baidu Cloud 9.4%"},"note":"T4 via 36Kr. Omdia's China AI cloud market for 2025 was RMB56.7B with Alibaba Cloud 38.1%, Volcano Engine 20.4%, Baidu Cloud 9.4%, Tencent Cloud 6.3% and eSurfing Cloud (China Telecom) 5.9%; the top five hold more than 80% (36Kr, 2026-08-04, citing Omdia data released 19 May; T4). Omdia's first-half 2025 report (SCMP via Yahoo Finance, 2025-09-10) defines the market as 'computing infrastructure optimised for generative AI tasks' and gave Alibaba 35.8%, Volcano Engine 14.8%, Huawei Cloud 13.1%, Tencent Cloud 7.0% and Baidu Cloud 6.1%. The full-year shares differ in structure (Volcano 14.8 to 20.4, Baidu 6.1 to 9.4, Huawei 13.1 to outside the top five), which suggests the full-year measure also counts model services, so every full-year China share is flagged offBasis; the first-half shares of the four providers named in both reports lie inside the widened ranges. Omdia gives no earlier full-year vendor shares, so there is no prior share on the same basis. Baidu's own results: AI Cloud Infra revenue RMB7.3B in Q2 2026 (+50%); GPU Cloud revenue grew 283% (no amount disclosed)."}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2},"china":{"levelLow":2,"levelHigh":4}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"21500951c1c993109c7d5007dd0360408442ea3db3c867cd54d6c6b72edb2d6d"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"21500951c1c993109c7d5007dd0360408442ea3db3c867cd54d6c6b72edb2d6d"}},"activityGrowth":{"current":{"low":7300000000,"central":7300000000,"high":7300000000,"unit":"RMB","basis":"Baidu AI Cloud Infra quarterly revenue, renminbi","period":"Q2 2026","periodEnd":"2026-06-30","source":{"url":"https://www.prnewswire.com/news-releases/baidu-announces-second-quarter-2026-results-302853860.html","title":"Baidu Announces Second Quarter 2026 Results","publisher":"Baidu, Inc.","publishedDate":"2026-08-18","accessedDate":"2026-10-22","tier":"T2","locator":"'Revenue from AI Cloud Infra was RMB 7.3 billion in the second quarter of 2026, up 50% year over year.'"},"note":"Company release; includes GPU Cloud (up 283% year on year, no amount disclosed) and other AI infrastructure."},"prior":{"low":4900000000,"central":4900000000,"high":4900000000,"unit":"RMB","basis":"Baidu AI Cloud Infra quarterly revenue, renminbi","period":"Q2 2025","periodEnd":"2025-06-30","source":{"url":"https://www.prnewswire.com/news-releases/baidu-announces-second-quarter-2026-results-302853860.html","title":"Baidu Announces Second Quarter 2026 Results","publisher":"Baidu, Inc.","publishedDate":"2026-08-18","accessedDate":"2026-10-22","tier":"T2","locator":"Q2 2025 AI Cloud Infra RMB 4.9 billion (summary table)"},"note":"Company release, comparison column."}},"baselineRevenueUsd":{"low":4303500000,"central":4303500000,"high":4303500000,"unit":"USD","basis":"Baidu AI Cloud Infra revenue, latest quarter x 4, US dollars","period":"Q2 2026 annualised","periodEnd":"2026-06-30","source":{"url":"https://www.prnewswire.com/news-releases/baidu-announces-second-quarter-2026-results-302853860.html","title":"Baidu Announces Second Quarter 2026 Results","publisher":"Baidu, Inc.","publishedDate":"2026-08-18","accessedDate":"2026-10-22","tier":"T2","locator":"AI Cloud Infra RMB 7.3 billion in Q2 2026 (RMB 8.8 billion in Q1 2026); RMB 6.7851 per US$1.00 at 30 June 2026"},"note":"RMB 29.2B converted at 6.7851 (the release's rate for 30 June 2026). The unit's own line, not Baidu's total. No share-change or committed-expansion input: Omdia gives no earlier share on the same basis and Baidu reports no backlog."}},"iren":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"iren","name":"IREN","kind":"public","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":0.327,"central":1.161,"high":1.857,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"ARR as of 26 Aug 2026 (company metric: GPU/hour pricing of commissioned GPUs x 8,760)","periodEnd":"2026-08-26","source":{"url":"https://www.sec.gov/Archives/edgar/data/0001878848/000187884826000051/irenreportsfy26results.htm","title":"IREN Reports FY26 Results (Form 8-K)","publisher":"IREN Limited","publishedDate":"2026-08-27","accessedDate":"2026-10-22","tier":"T5","locator":"'$1bn' operating ARR calculated as of August 26, 2026; AI Cloud Services revenue FY26 $128.8M; Q4 FY26 $70.5M"},"note":"Company metric, not recognised revenue: operating ARR $1bn calculated as of 26 August 2026 (release, 2026-08-27; commissioned GPUs at hourly pricing x 8,760 hours). Recognised AI Cloud revenue was $70.5M in the June 2026 quarter ($282M annualised) and $128.8M for FY2026; the company warns that recognised revenue may be materially lower than ARR. Central uses ARR; low is the recognised quarter annualised; high is the T5 upper width (+60%) above the ARR, $1.6B (contracted ARR of $4bn is for capacity due by 31 December 2026, so it is not current activity). 1,000 / 86,152. Denominator (the same in every file of this sublayer): GPU computing outside mainland China = about $86.2B a year, annualised from Q2 2026 (Oracle: June-August 2026). It is the sum of the provider estimates: AWS 11.0, Azure 18.0 and Google Cloud 9.0 (assumed splits of company-reported AI figures; none reports GPU rental), Oracle OCI 20.0, CoreWeave 10.3 (Q2 revenue $2,575M x 4), Nebius 2.3 (AI cloud revenue $574.9M x 4), all other registered providers 9.6, and 6.0 for providers outside the register. Reconciliation with published totals: Gartner forecasts AI-optimized IaaS spending of $42.3B for all of 2026 (Techstrong, 2026-09-21; Gartner's 10 August 2026 release is titled as a worldwide forecast; gartner.com returned HTTP 403 and was not retried, so its definition is unread). CoreWeave 10.3 + Nebius 2.3 + Oracle's CPU and GPU compute revenue of 26.0 ($6.5B in June-August 2026 x 4, The Next Platform, 2026-09-14) already total 38.6, 91% of Gartner's whole-year worldwide figure, before AWS, Microsoft, Google or any other provider. Gartner's category must therefore be narrower than AI compute rented to customers (or low), so it is not used as the denominator; it is the lowest case. Synergy's neocloud revenue above $25B for 2025 (2026-04-02) counts GPU-as-a-service, generative-AI platforms and high-density data-centre capacity and lists OpenAI and Anthropic among cloud providers, so it is wider than rented GPUs; the registered neoclouds' rental revenue sums to about 19.4 here. 5LayerCake judgement of the plausible range for the total: about $58B (every provider at the low end of its own range) to about $115B; the T5 share width covers it. Hyperscalers resell some capacity bought from neoclouds, so the total can double count by a few percent."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://iren.com/","title":"IREN home page","publisher":"IREN Limited","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Page: AI Cloud with NVIDIA reference architecture GPUs; no IREN-designed accelerator"}},"china":{"participates":false,"evidence":{"url":"https://iren.com/","title":"IREN home page","publisher":"IREN Limited","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Page lists sites in Canada (British Columbia) and Texas; it does not mention China"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"f6c60e8c69e4c36825969bbc4bc37f49247cb20486129c61745f4975b6cfbc95"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"f6c60e8c69e4c36825969bbc4bc37f49247cb20486129c61745f4975b6cfbc95"}},"activityGrowth":{"current":{"low":128800000,"central":128800000,"high":128800000,"unit":"USD","basis":"IREN AI Cloud revenue, fiscal year ended 30 June, US dollars","period":"FY2026","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0001878848/000187884826000051/irenreportsfy26results.htm","title":"IREN Reports FY26 Results (Form 8-K)","publisher":"IREN Limited","publishedDate":"2026-08-27","accessedDate":"2026-10-22","tier":"T1","locator":"AI Cloud revenue FY26 $128.8M"},"note":"Company results (Form 8-K)."},"prior":{"low":16400000,"central":16400000,"high":16400000,"unit":"USD","basis":"IREN AI Cloud revenue, fiscal year ended 30 June, US dollars","period":"FY2025","periodEnd":"2025-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0001878848/000187884826000051/irenreportsfy26results.htm","title":"IREN Reports FY26 Results (Form 8-K)","publisher":"IREN Limited","publishedDate":"2026-08-27","accessedDate":"2026-10-22","tier":"T1","locator":"AI Cloud revenue FY25 $16.4M ('~8x to $128.8m in FY26')"},"note":"Company results (Form 8-K), comparison column."}},"committedExpansion":{"low":4,"central":4,"high":4,"unit":"ratio","basis":"IREN contracted AI Cloud ARR for 2026 capacity ÷ operating AI Cloud ARR today","period":"ARR as of 26 August 2026","periodEnd":"2026-08-26","source":{"url":"https://www.sec.gov/Archives/edgar/data/0001878848/000187884826000051/irenreportsfy26results.htm","title":"IREN Reports FY26 Results (Form 8-K)","publisher":"IREN Limited","publishedDate":"2026-08-27","accessedDate":"2026-10-22","tier":"T2","locator":"'$4bn Contracted ARR for 2026 Capacity; $1bn Operating ARR Today.'"},"note":"The company says: contracted ARR $4bn for capacity targeted to be operational by 31 December 2026 and operating ARR $1bn today; 4.0 / 1.0 = 4.0. Both are the company's ARR metric, not recognised revenue; the contracted figure includes the operating $1bn."},"baselineRevenueUsd":{"low":128800000,"central":128800000,"high":128800000,"unit":"USD","basis":"IREN AI Cloud revenue, fiscal year ended 30 June, US dollars","period":"FY2026","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0001878848/000187884826000051/irenreportsfy26results.htm","title":"IREN Reports FY26 Results (Form 8-K)","publisher":"IREN Limited","publishedDate":"2026-08-27","accessedDate":"2026-10-22","tier":"T1","locator":"AI Cloud revenue FY26 $128.8M"},"note":"The AI Cloud line only (total revenue $707.0M, of which bitcoin mining $578.2M). No prior-share input: no same-basis share for mid-2025 was found."}},"tencent-cloud":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"tencent-cloud","name":"Tencent Cloud","kind":"unit","parentId":"tencent","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":0.174,"central":0.174,"high":0.174,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"2026 (page undated); revenue estimated","periodEnd":"2026-10-22","source":{"url":"https://www.tencentcloud.com/","title":"Tencent Cloud international site","publisher":"Tencent Cloud","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"International site lists 14 locations outside mainland China and a Cloud GPU Service; no scale"},"note":"5LayerCake T5 estimate; Tencent does not report cloud or GPU rental revenue. Assumed overseas GPU rental $0.15B. 150 / market total. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://www.tencentcloud.com/","title":"Tencent Cloud international site","publisher":"Tencent Cloud","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Page lists a Cloud GPU Service; no Tencent-designed accelerator named"}},"china":{"low":6.3,"central":6.3,"high":6.3,"unit":"percent","basis":"Share of AI-cloud market revenue in mainland China, calendar year 2025 (Omdia; includes model-as-a-service and platform revenue, wider than rented computing)","offBasis":true,"period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://eu.36kr.com/en/p/3924679520647302","title":"B-End Market Saturation Reached: Untapped C-End Frontier Becomes New Competitive Battlefield for Cloud Vendors to Seize Market Territory (citing Omdia, China AI cloud market 2025)","publisher":"36Kr","publishedDate":"2026-08-04","accessedDate":"2026-10-22","tier":"T4","locator":"Omdia China AI cloud market 2025: Tencent Cloud 6.3%"},"note":"T4 via 36Kr. Omdia's China AI cloud market for 2025 was RMB56.7B with Alibaba Cloud 38.1%, Volcano Engine 20.4%, Baidu Cloud 9.4%, Tencent Cloud 6.3% and eSurfing Cloud (China Telecom) 5.9%; the top five hold more than 80% (36Kr, 2026-08-04, citing Omdia data released 19 May; T4). Omdia's first-half 2025 report (SCMP via Yahoo Finance, 2025-09-10) defines the market as 'computing infrastructure optimised for generative AI tasks' and gave Alibaba 35.8%, Volcano Engine 14.8%, Huawei Cloud 13.1%, Tencent Cloud 7.0% and Baidu Cloud 6.1%. The full-year shares differ in structure (Volcano 14.8 to 20.4, Baidu 6.1 to 9.4, Huawei 13.1 to outside the top five), which suggests the full-year measure also counts model services, so every full-year China share is flagged offBasis; the first-half shares of the four providers named in both reports lie inside the widened ranges. Omdia gives no earlier full-year vendor shares, so there is no prior share on the same basis. Tencent does not report cloud revenue separately (Q2 2026: FinTech and Business Services RMB60.3B, up 9%, mainly cloud); no baseline, growth or backlog input exists."}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2},"china":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"a0dd9173339dc0eabf3bb7f812106e1fa3d8967201db9882c28c12c54bdbadcf"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"a0dd9173339dc0eabf3bb7f812106e1fa3d8967201db9882c28c12c54bdbadcf"}}},"huawei-cloud":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"huawei-cloud","name":"Huawei Cloud","kind":"unit","parentId":"huawei","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":0.116,"central":0.116,"high":0.116,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"2026 (page undated); revenue estimated","periodEnd":"2026-10-22","source":{"url":"https://www.huaweicloud.com/intl/en-us/about/index.html","title":"Why Huawei Cloud (international site)","publisher":"Huawei Cloud","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"34 regions, 103 availability zones, services in 170+ countries; 3 liquid-cooled AI computing centres; the page names no GPU or Ascend rental product, revenue or scale"},"note":"5LayerCake T5 estimate; Huawei does not report overseas AI rental and no page opened gives its GPU scale or revenue, so about $0.1B is assumed (a guess, unverified for this function; the company's activity in the China function is shown by Omdia's first-half 2025 report). The international ModelArts page (https://www.huaweicloud.com/intl/en-us/product/modelarts.html, opened 2026-10-22) offers 'Cost-effective Ascend computing power' and links to an ap-southeast-1 console, so Huawei probably also rents its own Ascend chips outside China; that is recorded in custom-accelerator-capacity as a small T5 allowance, not here. 100 / 86,152. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate."},"custom-accelerator-capacity":{"low":0.109,"central":0.543,"high":1.63,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"2026; revenue estimated","periodEnd":"2026-10-22","source":{"url":"https://www.huaweicloud.com/intl/en-us/product/modelarts.html","title":"ModelArts AI Development Platform | Huawei Cloud","publisher":"Huawei Cloud","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"International ModelArts page: 'Cost-effective Ascend computing power'; the Buy, Console and After-Sales links use region=ap-southeast-1; the page gives no capacity, customers or revenue (no date shown; footer '© 2026')"},"note":"5LayerCake T5 allowance (judgement, unverified): the opened international ModelArts page offers Huawei's own Ascend computing power from a Singapore (ap-southeast-1) console, so some Ascend capacity is probably sold to customers outside mainland China, but no page gives its scale. Assumed $50M a year, half of the $0.1B assumed for Huawei's rented GPUs outside China (gpu-capacity), with a range of $10M to $150M. 50 / 9,200 = 0.543%; 10 / 9,200 = 0.109%; 150 / 9,200 = 1.630%. Denominator: rented computing on cloud-owned AI chips outside China = about $9.2B a year (see the aws.json and google-cloud.json notes: Google 4.1, AWS 4.1, other custom-chip clouds 1.0); Huawei's $50M sits inside the 'other' 1.0, so the Google and AWS shares do not move. Ascend capacity rented inside mainland China is counted only in the China function, not here (no double count). Replaces the earlier 'not participating' entry, which rested on an overview page that does not mention Ascend. Export controls may limit overseas Ascend supply (Gartner, via SDxCentral, 2025-08-15), which is why the central is small."},"china":{"low":2,"central":4.5,"high":5.9,"unit":"percent","basis":"Share of AI-cloud market revenue in mainland China, calendar year 2025 (Omdia; includes model-as-a-service and platform revenue, wider than rented computing)","offBasis":true,"period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://eu.36kr.com/en/p/3924679520647302","title":"B-End Market Saturation Reached: Untapped C-End Frontier Becomes New Competitive Battlefield for Cloud Vendors to Seize Market Territory (citing Omdia, China AI cloud market 2025)","publisher":"36Kr","publishedDate":"2026-08-04","accessedDate":"2026-10-22","tier":"T5","locator":"Omdia China AI cloud market 2025 (RMB56.7B): the article names Alibaba Cloud 38.1%, Volcano Engine 20.4%, Baidu Cloud 9.4%, Tencent Cloud 6.3% and eSurfing Cloud 5.9% as the top five ('more than 80%'); Huawei Cloud is not named, so its share is below 5.9%"},"note":"5LayerCake T5 judgement under a T4 ceiling. Ceiling: Omdia's full-year 2025 top five in China's AI cloud market (RMB56.7B) are Alibaba Cloud 38.1%, Volcano Engine 20.4%, Baidu Cloud 9.4%, Tencent Cloud 6.3% and eSurfing Cloud (China Telecom) 5.9%, together more than 80% (36Kr, 2026-08-04, citing Omdia data released 19 May 2026; T4). Huawei Cloud is not named, so its share is below eSurfing's 5.9% and the high end is set at 5.9. The central 4.5 and low 2.0 are 5LayerCake judgement: the five named hold 80.1%, the other 19.9% is shared by every other provider, and no source gives Huawei's figure; the range allows for Huawei being among the larger or only a middling part of that remainder. Conflicting observation, outside the range: Omdia's first-half 2025 report gave Huawei Cloud 13.1%, third (SCMP via Yahoo Finance, 2025-09-10, https://finance.yahoo.com/news/alibaba-holds-wide-lead-over-093000369.html, T4; defined as 'computing infrastructure optimised for generative AI tasks'). IT Home (2026-05-19, https://www.ithome.com/0/952/224.htm) reports Alibaba's full-year 38.1% as up from 35.8% in the first half, the figure the first-half report gave, so the two reports measure the same market and a first-half third place cannot stand beside a full-year position outside the top five; the later, complete full-year list is used as the ceiling. IT Home also reports that Omdia defines the market as AI IaaS (RMB39.2B, 69%) plus model-as-a-service (RMB17.5B, 31%), which is wider than rented computing, so the observation is flagged offBasis. Omdia gives no earlier full-year vendor shares, so there is no prior share on the same basis. Huawei Cloud's external cloud revenue fell 3.5% in 2025 to RMB32.16B (CNBC via TrendForce, 2026-04-01)."}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2},"china":{"levelLow":2,"levelHigh":4},"custom-accelerator-capacity":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"ff1ae8ca0a27d3c615f20abd7eb1da53f1266fcac46f48d7e48f67c7495f5e0c"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"ff1ae8ca0a27d3c615f20abd7eb1da53f1266fcac46f48d7e48f67c7495f5e0c"}},"baselineRevenueUsd":{"low":4600000000,"central":4600000000,"high":4600000000,"unit":"USD","basis":"Huawei Cloud computing revenue from external customers, calendar year, US dollars","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.trendforce.com/news/2026/04/01/news-huawei-2025-r-cloud-revenue-slips/","title":"Huawei 2025 cloud revenue slips (citing CNBC and Reuters)","publisher":"TrendForce","publishedDate":"2026-04-01","accessedDate":"2026-10-22","tier":"T3","locator":"'cloud computing revenue from external customers fell 3.5% in 2025 to 32.16 billion yuan' ($4.6 billion), CNBC via TrendForce"},"note":"The unit's external cloud revenue as reported (RMB32.16B, $4.6B in the article); wider than AI compute. Segment of the unit, not Huawei's total."}},"china-telecom-tianyi-cloud":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"china-telecom-tianyi-cloud","name":"China Telecom (Tianyi Cloud)","kind":"unit","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":0.046,"central":0.046,"high":0.046,"period":"2026; overseas rental estimated","periodEnd":"2026-10-22","source":{"url":"https://www1.hkexnews.hk/listedco/listconews/sehk/2026/0820/2026082000375.pdf","title":"China Telecom Corporation Limited: Announcement of Interim Results for the six months ended 30 June 2026","publisher":"China Telecom Corporation Limited","publishedDate":"2026-08-20","accessedDate":"2026-10-22","tier":"T5","locator":"H1 2026: 'revenue from intelligent computing increased by 95.0% year-on-year'; proprietary intelligent computing capacity 70.8 EFLOPS; 'Four heterogeneous computing power clusters, each with 10,000 GPUs'; utilisation 94%; 'kept expanding overseas markets' with no figures"},"note":"5LayerCake T5 estimate. The interim announcement shows GPU computing activity in China and says the company 'kept expanding overseas markets', but gives no overseas GPU rental figure, so about $0.04B outside China is assumed (a guess). 40 / 86,152. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate.","unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars"},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://www1.hkexnews.hk/listedco/listconews/sehk/2026/0820/2026082000375.pdf","title":"China Telecom Corporation Limited: Announcement of Interim Results for the six months ended 30 June 2026","publisher":"China Telecom Corporation Limited","publishedDate":"2026-08-20","accessedDate":"2026-10-22","tier":"T1","locator":"Xirang platform supports 'more than 20 types of heterogeneous chips'; the announcement names no China Telecom-designed accelerator offered for rent abroad"}},"china":{"low":5.9,"central":5.9,"high":5.9,"unit":"percent","basis":"Share of AI-cloud market revenue in mainland China, calendar year 2025 (Omdia; includes model-as-a-service and platform revenue, wider than rented computing)","offBasis":true,"period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://eu.36kr.com/en/p/3924679520647302","title":"B-End Market Saturation Reached: Untapped C-End Frontier Becomes New Competitive Battlefield for Cloud Vendors to Seize Market Territory (citing Omdia, China AI cloud market 2025)","publisher":"36Kr","publishedDate":"2026-08-04","accessedDate":"2026-10-22","tier":"T4","locator":"Omdia China AI cloud market 2025: eSurfing Cloud 5.9%"},"note":"T4 via 36Kr. Omdia's China AI cloud market for 2025 was RMB56.7B with Alibaba Cloud 38.1%, Volcano Engine 20.4%, Baidu Cloud 9.4%, Tencent Cloud 6.3% and eSurfing Cloud (China Telecom) 5.9%; the top five hold more than 80% (36Kr, 2026-08-04, citing Omdia data released 19 May; T4). Omdia's first-half 2025 report (SCMP via Yahoo Finance, 2025-09-10) defines the market as 'computing infrastructure optimised for generative AI tasks' and gave Alibaba 35.8%, Volcano Engine 14.8%, Huawei Cloud 13.1%, Tencent Cloud 7.0% and Baidu Cloud 6.1%. The full-year shares differ in structure (Volcano 14.8 to 20.4, Baidu 6.1 to 9.4, Huawei 13.1 to outside the top five), which suggests the full-year measure also counts model services, so every full-year China share is flagged offBasis; the first-half shares of the four providers named in both reports lie inside the widened ranges. Omdia gives no earlier full-year vendor shares, so there is no prior share on the same basis. China Telecom's cloud business revenue was RMB61.8B in the first half of 2026 (up 7.8%) and its 'intelligence' business RMB31.1B (up 7.1%); neither is AI-compute-only."}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2},"china":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"a694f948a194bfcfd3f7b9566caa383b39abb9dd7ad15871df7c17cceae573d5"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"a694f948a194bfcfd3f7b9566caa383b39abb9dd7ad15871df7c17cceae573d5"}},"baselineRevenueUsd":{"low":18216400000,"central":18216400000,"high":18216400000,"unit":"USD","basis":"China Telecom Cloud business revenue, first half x 2, US dollars","period":"H1 2026 annualised","periodEnd":"2026-06-30","source":{"url":"https://www1.hkexnews.hk/listedco/listconews/sehk/2026/0820/2026082000375.pdf","title":"China Telecom Corporation Limited: Announcement of Interim Results for the six months ended 30 June 2026","publisher":"China Telecom Corporation Limited","publishedDate":"2026-08-20","accessedDate":"2026-10-22","tier":"T1","locator":"'Revenue from China Telecom Cloud reached RMB61.8 billion, representing an increase of 7.8% year-on-year' (H1 2026); Intelligence Business RMB31.1 billion (+7.1%)"},"note":"RMB61.8B x 2 = RMB123.6B at RMB 6.7851 per US$ (the noon buying rate on 30 June 2026 certified by the Federal Reserve Bank of New York, as printed in Kingsoft Cloud's Q2 2026 release, https://www.sec.gov/Archives/edgar/data/0001795589/000110465926098490/tm2623439d2_ex99-1.htm) = US$18.2B. The whole cloud business, wider than AI compute; read from the company's interim results announcement (HKEX); The Standard (Hong Kong, 2026-08-20) reports the same figure."}},"crusoe":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"crusoe","name":"Crusoe","kind":"private","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":0.929,"central":0.929,"high":0.929,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"CY2026 expected (Newcomer, Sep 2026)","periodEnd":"2026-09-23","source":{"url":"https://www.newcomer.co/p/crusoe-is-on-track-for-2-billion","title":"Crusoe is on track for $2 billion","publisher":"Newcomer","publishedDate":"2026-09-23","accessedDate":"2026-10-22","tier":"T5","locator":"'on track for $2 billion in revenue this year' (Newcomer, 23 Sep 2026; paywalled, summary read)"},"note":"Newcomer reports Crusoe is on track for $2B revenue in 2026 but the cloud/data-centre split sits behind a paywall. Crusoe's Series F release (2026-09-17) gives no revenue: it reports $140B+ total contracted value across the platform, over 6 GW of contracted capacity, 1 GW delivered, 20x+ growth in Crusoe Cloud bookings and over $100M contracted ARR for Managed Inference (per-token, so Inference serving). Assumed cloud rental about 40% of $2B = $0.8B. 800 / 86,152. No Rising input could be sourced: no cloud-only revenue for two periods, and the $140B backlog mixes campus and cloud contracts. Denominator (the same in every file of this sublayer): GPU computing outside mainland China = about $86.2B a year, annualised from Q2 2026 (Oracle: June-August 2026). It is the sum of the provider estimates: AWS 11.0, Azure 18.0 and Google Cloud 9.0 (assumed splits of company-reported AI figures; none reports GPU rental), Oracle OCI 20.0, CoreWeave 10.3 (Q2 revenue $2,575M x 4), Nebius 2.3 (AI cloud revenue $574.9M x 4), all other registered providers 9.6, and 6.0 for providers outside the register. Reconciliation with published totals: Gartner forecasts AI-optimized IaaS spending of $42.3B for all of 2026 (Techstrong, 2026-09-21; Gartner's 10 August 2026 release is titled as a worldwide forecast; gartner.com returned HTTP 403 and was not retried, so its definition is unread). CoreWeave 10.3 + Nebius 2.3 + Oracle's CPU and GPU compute revenue of 26.0 ($6.5B in June-August 2026 x 4, The Next Platform, 2026-09-14) already total 38.6, 91% of Gartner's whole-year worldwide figure, before AWS, Microsoft, Google or any other provider. Gartner's category must therefore be narrower than AI compute rented to customers (or low), so it is not used as the denominator; it is the lowest case. Synergy's neocloud revenue above $25B for 2025 (2026-04-02) counts GPU-as-a-service, generative-AI platforms and high-density data-centre capacity and lists OpenAI and Anthropic among cloud providers, so it is wider than rented GPUs; the registered neoclouds' rental revenue sums to about 19.4 here. 5LayerCake judgement of the plausible range for the total: about $58B (every provider at the low end of its own range) to about $115B; the T5 share width covers it. Hyperscalers resell some capacity bought from neoclouds, so the total can double count by a few percent."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://www.crusoe.ai/cloud","title":"Crusoe Cloud product page","publisher":"Crusoe","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Cloud page lists NVIDIA (GB300, B300, H200, H100) and AMD (MI300X, MI355X) GPUs; no Crusoe-designed accelerator"}},"china":{"participates":false,"evidence":{"url":"https://www.crusoe.ai/cloud","title":"Crusoe Cloud product page","publisher":"Crusoe","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Page does not mention China; no data-centre locations named"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"abeb5e9acbc62c4d33f3a69d2ae65639f136042242084ab9af1dcc5e8a14ea68"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"abeb5e9acbc62c4d33f3a69d2ae65639f136042242084ab9af1dcc5e8a14ea68"}}},"together-ai":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"together-ai","name":"Together AI","kind":"private","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":0.754,"central":0.754,"high":0.754,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"February 2026 annualised","periodEnd":"2026-02-28","source":{"url":"https://sacra.com/c/together-ai/","title":"Together AI (company research page)","publisher":"Sacra","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Sacra: about $1B annualised revenue in February 2026; per-token API roughly 30-40% of revenue, GPU server rentals most of the rest; annual bookings topped $1.15B (Q2 2026)"},"note":"T4 estimate by Sacra (about $1B annualised revenue, February 2026) with GPU rentals about 60-70% (the page gives no exact figure); 5LayerCake central $0.65B for rental. Per-token API revenue belongs to Models - Inference serving. Together's own Series C post (2026-07-01) discloses no revenue. 650 / market total. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://www.together.ai/","title":"Together AI home page","publisher":"Together AI","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Home page: self-serve instant clusters of NVIDIA GPUs (B200); no Together-designed accelerator"}},"china":{"participates":false,"evidence":{"url":"https://www.together.ai/","title":"Together AI home page","publisher":"Together AI","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Page does not mention China or any China site"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"6e9432008e5368421563cf0a39a1aabae564341280d2a12ccfb3da63ca731133"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"6e9432008e5368421563cf0a39a1aabae564341280d2a12ccfb3da63ca731133"}},"baselineRevenueUsd":{"low":1000000000,"central":1000000000,"high":1000000000,"unit":"USD","basis":"Together AI total revenue, annualised, US dollars","period":"February 2026 annualised","periodEnd":"2026-02-28","source":{"url":"https://sacra.com/c/together-ai/","title":"Together AI (company research page)","publisher":"Sacra","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T4","locator":"Sacra: about $1B annualised revenue in February 2026"},"note":"Sacra estimate (T4); includes per-token API revenue (about 30-40%), which belongs to Inference serving."}},"kingsoft-cloud":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"kingsoft-cloud","name":"Kingsoft Cloud","kind":"public","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":0.046,"central":0.046,"high":0.046,"period":"2026; overseas rental estimated","periodEnd":"2026-10-22","source":{"url":"https://www.sec.gov/Archives/edgar/data/0001795589/000110465926098490/tm2623439d2_ex99-1.htm","title":"Kingsoft Cloud Announces Unaudited Second Quarter 2026 Financial Results (Form 6-K exhibit 99.1)","publisher":"Kingsoft Cloud Holdings Limited","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Q2 2026: 'AI cloud gross billings reached RMB1,327 million (US$195.6 million)', +82% YoY; growth 'driven by incremental contributions from our AI cloud infrastructure services'; no GPU count and no overseas revenue is given (the release carries no dateline; the earnings call it announces was on 19 August 2026)"},"note":"5LayerCake T5 estimate: the release shows AI cloud infrastructure activity in China but no overseas GPU rental, so about $40M a year outside China is assumed (a guess). 40 / 86,152. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate.","unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars"},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/0001795589/000110465926098490/tm2623439d2_ex99-1.htm","title":"Kingsoft Cloud Announces Unaudited Second Quarter 2026 Financial Results (Form 6-K exhibit 99.1)","publisher":"Kingsoft Cloud Holdings Limited","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Release cites 'continued investment in AI computing resources' and 'investment into the procurement of computing power equipment'; no Kingsoft-designed accelerator offered for rent is mentioned"}},"china":{"low":3,"central":4.5,"high":5.9,"unit":"percent","basis":"Share of AI-cloud market revenue in mainland China, calendar year 2025 (Omdia; includes model-as-a-service and platform revenue, wider than rented computing)","offBasis":true,"period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www.sec.gov/Archives/edgar/data/1795589/000110465926034141/tm269536d1_ex99-1.htm","title":"Kingsoft Cloud Announces Unaudited Fourth Quarter and Fiscal Year 2025 Financial Results (Form 6-K exhibit 99.1)","publisher":"Kingsoft Cloud Holdings Limited","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Q4 2025: 'The gross billing of AI business achieved RMB926 million this quarter' (+95% YoY); with the Q1-Q3 2025 releases listed in the note, FY2025 AI gross billing is RMB2,962.1M (the release carries no dateline; the earnings call it announces was on 25 March 2026)"},"note":"5LayerCake T5 derivation under a T4 ceiling. Kingsoft Cloud's AI business gross billing in 2025 was RMB525M in Q1 (https://www.sec.gov/Archives/edgar/data/1795589/000110465925053312/tm2516063d1_ex99-1.htm), RMB728.7M in Q2 (https://www.sec.gov/Archives/edgar/data/1795589/000110465925080532/tm2523624d1_ex99-1.htm), RMB782.4M in Q3 (https://www.sec.gov/Archives/edgar/data/1795589/000110465925113842/tm2530606d1_ex99-1.htm) and RMB926M in Q4 (the cited release), a total of RMB2,962.1M (company releases, T2). Divided by Omdia's RMB56.7B China AI cloud market for 2025 (36Kr, T4) that is 5.2%. Gross billing is the company's AI-business billing, not recognised revenue, and spans AI cloud infrastructure and model services, while Omdia measures market revenue, so 5.2% is treated as an upper estimate; the central 4.5% and low 3.0% allow for billing exceeding recognised revenue (5LayerCake judgement). Omdia's full-year 2025 list of the top five in China's AI cloud market (RMB56.7B) is Alibaba Cloud 38.1%, Volcano Engine 20.4%, Baidu Cloud 9.4%, Tencent Cloud 6.3% and eSurfing Cloud 5.9%, together more than 80% (36Kr, 2026-08-04, citing Omdia data released 19 May 2026; T4); Kingsoft Cloud is not named, so its share is below 5.9%, which is the high end. IT Home (2026-05-19, https://www.ithome.com/0/952/224.htm) reports that Omdia defines the market as AI IaaS (RMB39.2B, 69%) plus model-as-a-service (RMB17.5B, 31%), which is wider than rented computing, so every China share is flagged offBasis. Omdia gives no earlier full-year vendor shares, so there is no prior share on the same basis."}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2},"china":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"a4d568dd76dff7c9602bb8a700467c9a40823b55428baad29b6f3605623f0ca4"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"a4d568dd76dff7c9602bb8a700467c9a40823b55428baad29b6f3605623f0ca4"}}},"core42":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"core42","name":"Core42 (G42)","kind":"unit","parentId":"g42","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":0.58,"central":0.58,"high":0.58,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"April 2026 fleet, annualised revenue estimated","periodEnd":"2026-04-30","source":{"url":"https://www.core42.ai/resources/whitepapers/core42-ai-cloud-your-gateway-to-the-future-of-ai","title":"Core42 AI Cloud: your gateway to the future of AI (whitepaper, April 2026)","publisher":"Core42","publishedDate":"2026-04","accessedDate":"2026-10-22","tier":"T5","locator":"'86K+ GPUs' across sovereign data centres (April 2026 whitepaper); MBZUAI case study"},"note":"5LayerCake T5 estimate. Core42 says its AI Cloud operates 86K+ GPUs; it discloses no revenue. 86,000 GPUs x $11k per GPU-year (anchor: Bitdeer 6-K and HIVE show $6k-$18k) x about 55% rented utilisation = $0.5B. 500 / market total. Part of the fleet serves G42's own use, so the figure may be high. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://www.core42.ai/resources/whitepapers/core42-ai-cloud-your-gateway-to-the-future-of-ai","title":"Core42 AI Cloud: your gateway to the future of AI (whitepaper, April 2026)","publisher":"Core42","publishedDate":"2026-04","accessedDate":"2026-10-22","tier":"T2","locator":"Whitepaper: 'GPU as a Service ... a diverse range of accelerators via bare metal, Kubernetes or Slurm'; no Core42-designed accelerator named"}},"china":{"participates":false,"evidence":{"url":"https://www.core42.ai/resources/whitepapers/core42-ai-cloud-your-gateway-to-the-future-of-ai","title":"Core42 AI Cloud: your gateway to the future of AI (whitepaper, April 2026)","publisher":"Core42","publishedDate":"2026-04","accessedDate":"2026-10-22","tier":"T2","locator":"Whitepaper lists sites in the US, UAE, Southern Europe, with Kenya, India and Southeast Asia in deployment; none in mainland China"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"365f3d72e4c52ca91a0e5fd884e5144df5eca88e48a7bdb6733b369063897734"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"365f3d72e4c52ca91a0e5fd884e5144df5eca88e48a7bdb6733b369063897734"}}},"china-mobile-cloud":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"china-mobile-cloud","name":"China Mobile Cloud","kind":"unit","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":0.035,"central":0.035,"high":0.035,"period":"2026; overseas rental estimated","periodEnd":"2026-10-22","source":{"url":"https://www1.hkexnews.hk/listedco/listconews/sehk/2026/0813/2026081300218.pdf","title":"China Mobile Limited: 2026 Interim Results (announcement)","publisher":"China Mobile Limited","publishedDate":"2026-08-13","accessedDate":"2026-10-22","tier":"T5","locator":"H1 2026: 'computing service revenue reached RMB52.9 billion' (+14.0%); 'Revenue from intelligent computing services reached RMB5.3 billion' (+130.1%); 'international market revenue reached RMB18.2 billion' is a telecommunications line; no overseas GPU rental is reported"},"note":"5LayerCake T5 estimate: the company's intelligent-computing revenue is mainland-focused and it discloses no overseas GPU rental, so about $30M a year outside China is assumed (a guess, unverified). The interim announcement (13 August 2026) shows activity in the sublayer in China, not abroad. 30 / 86,152. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate.","unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars"},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://www1.hkexnews.hk/listedco/listconews/sehk/2026/0813/2026081300218.pdf","title":"China Mobile Limited: 2026 Interim Results (announcement)","publisher":"China Mobile Limited","publishedDate":"2026-08-13","accessedDate":"2026-10-22","tier":"T1","locator":"Note 3 defines computing services as 'data centers, cloud computing services and cloud computing applications'; the announcement names no China Mobile-designed accelerator offered for rent"}},"china":{"low":3,"central":3,"high":3,"unit":"percent","basis":"Share of AI-cloud market revenue in mainland China, calendar year 2025 (Omdia; includes model-as-a-service and platform revenue, wider than rented computing)","offBasis":true,"period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www1.hkexnews.hk/listedco/listconews/sehk/2026/0813/2026081300218.pdf","title":"China Mobile Limited: 2026 Interim Results (announcement)","publisher":"China Mobile Limited","publishedDate":"2026-08-13","accessedDate":"2026-10-22","tier":"T5","locator":"H1 2026: 'public cloud revenue ranking among the top three in the domestic market for the first time' (IDC China Public Cloud Services Market Tracker, 2H 2025); 'Revenue from intelligent computing services reached RMB5.3 billion' (+130.1%). China Mobile is not among Omdia's full-year top five (36Kr)"},"note":"5LayerCake T5 judgement under a T4 ceiling. Omdia's full-year 2025 list of the top five in China's AI cloud market (RMB56.7B) is Alibaba Cloud 38.1%, Volcano Engine 20.4%, Baidu Cloud 9.4%, Tencent Cloud 6.3% and eSurfing Cloud 5.9%, together more than 80% (36Kr, 2026-08-04, citing Omdia data released 19 May 2026; T4). China Mobile Cloud is not named, so its share is below 5.9%. The 3.0% is a 5LayerCake allocation of the 19.9% left for all other providers (together with Huawei Cloud 4.5, Kingsoft Cloud 4.5, China Unicom 1.8, SenseTime 1.5, JD Cloud 1.0, UCloud 0.5 and Capital Online 0.2); no Omdia figure exists for it. The company's own lines are not on Omdia's basis and cannot be divided by Omdia's market: the interim announcement gives intelligent computing services revenue of RMB5.3B in H1 2026 (+130.1%, so about RMB2.3B in H1 2025), a company-defined line whose make-up it does not give, and computing service revenue of RMB52.9B (data centres, cloud computing services and cloud computing applications). IT Home (2026-05-19, https://www.ithome.com/0/952/224.htm) reports that Omdia defines the market as AI IaaS (RMB39.2B, 69%) plus model-as-a-service (RMB17.5B, 31%), which is wider than rented computing, so every China share is flagged offBasis."}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2},"china":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"7488ae6a4a9a4c57add83d61c56a37e2f1c17d4fddc0e0ee5de41c4cb5893e7f"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"7488ae6a4a9a4c57add83d61c56a37e2f1c17d4fddc0e0ee5de41c4cb5893e7f"}}},"fluidstack":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"fluidstack","name":"Fluidstack","kind":"private","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":0.522,"central":0.522,"high":0.522,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"Fleet 2026, annualised revenue estimated","periodEnd":"2026-04-30","source":{"url":"https://sacra.com/c/fluidstack","title":"Fluidstack (company research page)","publisher":"Sacra","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Sacra: manages more than 100,000 GPUs; $180M ARR at December 2024 (up 620%); Anthropic $50B partnership"},"note":"5LayerCake T5 estimate. Fluidstack is private; the only revenue datum read is Sacra's $180M ARR at December 2024 (stale and low). It manages more than 100,000 GPUs (partly owned by hosts and backed by Google). 100,000 x $9k per GPU-year x about 50% in service = $0.45B. 450 / market total. Large uncertainty. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://sacra.com/c/fluidstack","title":"Fluidstack (company research page)","publisher":"Sacra","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Sacra page: private cloud and marketplace of GPU clusters hosted at TeraWulf, Cipher and Hut 8 sites in the US; no Fluidstack-designed accelerator"}},"china":{"participates":false,"evidence":{"url":"https://sacra.com/c/fluidstack","title":"Fluidstack (company research page)","publisher":"Sacra","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Sacra page lists US sites (New York, Texas, Louisiana) and UK headquarters; no mainland China site"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"40b510c92788bc6e8bfd72cf175d5f2be000c50548352381f1a88496a62686f2"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"40b510c92788bc6e8bfd72cf175d5f2be000c50548352381f1a88496a62686f2"}}},"modal":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"modal","name":"Modal","kind":"private","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":0.522,"central":0.522,"high":0.522,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"October 2026 annualised","periodEnd":"2026-10-14","source":{"url":"https://modal.com/blog/1-billion-annualized-run-rate-series-d","title":"Modal crosses $1B in Annualized Run Rate and raises $750M Series D","publisher":"Modal","publishedDate":"2026-10-14","accessedDate":"2026-10-22","tier":"T5","locator":"'recently crossed $1B in annualized run rate'; Sandboxes 'a $300M run rate product'; capacity pool 'spans 20+ clouds'"},"note":"Company statement: annualised run rate above $1B, of which Sandboxes $300M. Assumed about 65% of the remaining $700M is GPU compute (the rest is CPU, storage and orchestration): $0.45B. Modal runs on capacity rented from 20+ clouds, so its revenue partly double counts the clouds. Per-second inference APIs overlap Inference serving. 450 / market total. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://modal.com/blog/1-billion-annualized-run-rate-series-d","title":"Modal crosses $1B in Annualized Run Rate and raises $750M Series D","publisher":"Modal","publishedDate":"2026-10-14","accessedDate":"2026-10-22","tier":"T2","locator":"Page: capacity pool of 20+ clouds; no Modal-designed accelerator"}},"china":{"participates":false,"evidence":{"url":"https://modal.com/blog/1-billion-annualized-run-rate-series-d","title":"Modal crosses $1B in Annualized Run Rate and raises $750M Series D","publisher":"Modal","publishedDate":"2026-10-14","accessedDate":"2026-10-22","tier":"T2","locator":"Page does not mention China; customers are US companies"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"7ba67301eea430386ed4eec0759a30b9fa1c95cd9363abecfd3058cf9e953966"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"7ba67301eea430386ed4eec0759a30b9fa1c95cd9363abecfd3058cf9e953966"}},"baselineRevenueUsd":{"low":1000000000,"central":1000000000,"high":1000000000,"unit":"USD","basis":"Modal annualised run-rate revenue, US dollars","period":"October 2026 annualised","periodEnd":"2026-10-14","source":{"url":"https://modal.com/blog/1-billion-annualized-run-rate-series-d","title":"Modal crosses $1B in Annualized Run Rate and raises $750M Series D","publisher":"Modal","publishedDate":"2026-10-14","accessedDate":"2026-10-22","tier":"T2","locator":"Modal: crossed $1B annualised run rate (Series D post, 14 Oct 2026)"},"note":"The company says it crossed $1B; includes Sandboxes ($300M) and CPU, storage and orchestration."}},"voltage-park":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"voltage-park","name":"Voltage Park","kind":"private","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":0.348,"central":0.348,"high":0.348,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"Fleet as shown 2026-10-22; revenue estimated","periodEnd":"2026-10-22","source":{"url":"https://www.voltagepark.com/","title":"Voltage Park home page","publisher":"Voltage Park","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"'36,000' NVIDIA Hopper, Blackwell and Grace Blackwell GPUs; on-demand from $1.99/hr; customers include Cursor, Phind, Luma AI"},"note":"5LayerCake T5 estimate. 36,000 GPUs x $9k per GPU-year x about 90% = $0.3B (anchor: Bitdeer shows $17.9k per GPU-year at 95% utilisation, HIVE $6.4k). Voltage Park publishes no revenue and the page is undated. 300 / market total. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://www.voltagepark.com/","title":"Voltage Park home page","publisher":"Voltage Park","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Page lists NVIDIA GPUs only"}},"china":{"participates":false,"evidence":{"url":"https://www.voltagepark.com/","title":"Voltage Park home page","publisher":"Voltage Park","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Page names no locations and does not mention China"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"eae8d65027ea76e6ddec1a4e3d0076bd88aa20d18339efc83e2bbf3f6058d856"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"eae8d65027ea76e6ddec1a4e3d0076bd88aa20d18339efc83e2bbf3f6058d856"}}},"nscale":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"nscale","name":"Nscale","kind":"private","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":0.325,"central":0.325,"high":0.325,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"H1 2026 annualised","periodEnd":"2026-06-30","source":{"url":"https://whbl.com/2026/09/18/ai-cloud-firm-nscale-files-for-us-ipo/","title":"AI cloud firm Nscale files for US IPO (Reuters)","publisher":"WHBL (Reuters wire)","publishedDate":"2026-09-18","accessedDate":"2026-10-22","tier":"T5","locator":"Revenue of $140.6M in the six months ended June 30, 2026 (vs $10.4M); S-1 filed 18 Sep 2026"},"note":"Reuters on Nscale's S-1: H1 2026 revenue $140.6M x 2 = $281M annualised. Revenue is ramping quickly (a call summary I did not open put Q2 above $100M), so the Q2 rate is higher than this H1 average. 280 / market total. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://www.nscale.com/","title":"Nscale home page","publisher":"Nscale","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Home page: on-demand GPU and bare metal, dedicated inference; no Nscale-designed accelerator"}},"china":{"participates":false,"evidence":{"url":"https://www.nscale.com/","title":"Nscale home page","publisher":"Nscale","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Page lists sites in Norway, UK, US, Portugal and Iceland; none in mainland China"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"6016d8c9ecbd7d3885efd736dc099aab1dab365691f6ffbfca212da0b3abfb14"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"6016d8c9ecbd7d3885efd736dc099aab1dab365691f6ffbfca212da0b3abfb14"}},"baselineRevenueUsd":{"low":281200000,"central":281200000,"high":281200000,"unit":"USD","basis":"Nscale revenue, first half x 2, US dollars","period":"H1 2026 annualised","periodEnd":"2026-06-30","source":{"url":"https://whbl.com/2026/09/18/ai-cloud-firm-nscale-files-for-us-ipo/","title":"AI cloud firm Nscale files for US IPO (Reuters)","publisher":"WHBL (Reuters wire)","publishedDate":"2026-09-18","accessedDate":"2026-10-22","tier":"T3","locator":"Reuters on the S-1: H1 2026 revenue $140.6M"},"note":"Reuters, citing the S-1 (WHBL wire copy). H1 average; the Q2 rate is higher."}},"china-unicom-cloud":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"china-unicom-cloud","name":"China Unicom Cloud","kind":"unit","commercialActivityVerified":false,"shares":{"gpu-capacity":{"low":0.023,"central":0.023,"high":0.023,"period":"2026; overseas rental estimated","periodEnd":"2026-10-22","source":{"url":"https://www1.hkexnews.hk/listedco/listconews/sehk/2026/0818/2026081800335.pdf","title":"China Unicom (Hong Kong) Limited: 2026 Interim Results Announcement","publisher":"China Unicom (Hong Kong) Limited","publishedDate":"2026-08-18","accessedDate":"2026-10-22","tier":"T5","locator":"H1 2026: 'Computing power revenue reaching RMB41.9 billion' (+13%), of which 'computing service revenue increased by 9%'; 'cross-region mixed training and inference on heterogeneous GPUs in large-scale production scenarios'; no overseas GPU rental is reported"},"note":"5LayerCake T5 estimate: the interim announcement shows GPU use and computing revenue in China but no overseas GPU rental (China Unicom Global Limited is named only as a subsidiary), so about $20M a year outside China is assumed (a guess). Unverified: the announcement's 'computing power revenue' (data centre, computing service, digital smart applications and cloud-AI services) is not AI compute rented to customers, so commercial activity in this sublayer is left unverified. 20 / 86,152. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate.","unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars"},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://www1.hkexnews.hk/listedco/listconews/sehk/2026/0818/2026081800335.pdf","title":"China Unicom (Hong Kong) Limited: 2026 Interim Results Announcement","publisher":"China Unicom (Hong Kong) Limited","publishedDate":"2026-08-18","accessedDate":"2026-10-22","tier":"T1","locator":"Note 3 defines computing power revenue as 'data centre revenue + computing service revenue + digital smart applications revenue + cloud-AI services revenue'; GPUs are described only as 'heterogeneous GPUs' and domestic intelligent computing power; no China Unicom-designed accelerator offered for rent is mentioned"}},"china":{"low":1.8,"central":1.8,"high":1.8,"unit":"percent","basis":"Share of AI-cloud market revenue in mainland China, calendar year 2025 (Omdia; includes model-as-a-service and platform revenue, wider than rented computing)","offBasis":true,"period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://www1.hkexnews.hk/listedco/listconews/sehk/2026/0818/2026081800335.pdf","title":"China Unicom (Hong Kong) Limited: 2026 Interim Results Announcement","publisher":"China Unicom (Hong Kong) Limited","publishedDate":"2026-08-18","accessedDate":"2026-10-22","tier":"T5","locator":"H1 2026: 'Computing power revenue reaching RMB41.9 billion' (+13%); the announcement gives no China Unicom Cloud revenue and no AI cloud market share. China Unicom is not among Omdia's full-year top five (36Kr)"},"note":"5LayerCake T5 judgement under a T4 ceiling. Omdia's full-year 2025 list of the top five in China's AI cloud market (RMB56.7B) is Alibaba Cloud 38.1%, Volcano Engine 20.4%, Baidu Cloud 9.4%, Tencent Cloud 6.3% and eSurfing Cloud 5.9%, together more than 80% (36Kr, 2026-08-04, citing Omdia data released 19 May 2026; T4). China Unicom Cloud is not named, so its share is below 5.9%. The 1.8% is a 5LayerCake allocation of the 19.9% left for all other providers (with Huawei Cloud 4.5, Kingsoft Cloud 4.5, China Mobile 3.0, SenseTime 1.5, JD Cloud 1.0, UCloud 0.5 and Capital Online 0.2); no Omdia figure exists for it. For scale only: the interim announcement's 'computing power revenue' of RMB41.9B (H1 2026, definition above) compares with China Mobile's computing service revenue of RMB52.9B, but the two lines are defined differently and neither is on Omdia's basis. IT Home (2026-05-19, https://www.ithome.com/0/952/224.htm) reports that Omdia defines the market as AI IaaS (RMB39.2B, 69%) plus model-as-a-service (RMB17.5B, 31%), which is wider than rented computing, so every China share is flagged offBasis."}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2},"china":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"aa5e0096fb0c8d2826909a6d00c2763c6fe66c390b2a8e8bfdb0edc5df92c4ec"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"aa5e0096fb0c8d2826909a6d00c2763c6fe66c390b2a8e8bfdb0edc5df92c4ec"}}},"runpod":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"runpod","name":"Runpod","kind":"private","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":0.279,"central":0.279,"high":0.279,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"June 2026 annualised","periodEnd":"2026-06-24","source":{"url":"https://thenextweb.com/news/runpod-100m-summit-partners-1bn-valuation","title":"Runpod hits $1bn valuation as the AI compute crunch bites","publisher":"The Next Web","publishedDate":"2026-06-24","accessedDate":"2026-10-22","tier":"T5","locator":"'doubled its annualised revenue to around $240M over the past five months' (The Information, as reported by TNW)"},"note":"Press report of a figure from The Information (T3); the company reports ARR of $325M in a release dated 13 Oct 2026 per a search summary I did not open. Runpod serverless and pods are rented GPU compute. 240 / market total. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://www.runpod.io/","title":"Runpod home page","publisher":"Runpod","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Home page: 30+ GPU SKUs from B200 to RTX 4090 (merchant GPUs)"}},"china":{"participates":false,"evidence":{"url":"https://www.runpod.io/","title":"Runpod home page","publisher":"Runpod","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Page cites 31 regions without naming them and does not mention China"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"5c62014c1208250721def511c0b6044e92647a64eca71d3373177448848a8609"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"5c62014c1208250721def511c0b6044e92647a64eca71d3373177448848a8609"}},"baselineRevenueUsd":{"low":240000000,"central":240000000,"high":240000000,"unit":"USD","basis":"Runpod annualised revenue, US dollars","period":"June 2026 annualised","periodEnd":"2026-06-30","source":{"url":"https://thenextweb.com/news/runpod-100m-summit-partners-1bn-valuation","title":"Runpod hits $1bn valuation as the AI compute crunch bites","publisher":"The Next Web","publishedDate":"2026-06-24","accessedDate":"2026-10-22","tier":"T3","locator":"The Next Web, citing The Information: about $240M"},"note":"Press report of a figure from The Information."}},"sensetime":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"sensetime","name":"SenseTime (SenseCore)","kind":"public","commercialActivityVerified":false,"shares":{"gpu-capacity":{"low":0.035,"central":0.035,"high":0.035,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"2026; overseas rental estimated","periodEnd":"2026-10-22","source":{"url":"https://www.sensetime.com/en","title":"SenseTime site (SenseCore AI Infrastructure)","publisher":"SenseTime","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Mainland-focused provider; overseas GPU rental not reported"},"note":"5LayerCake T5 estimate: overseas GPU rental is not disclosed and assumed negligible (about $30M). 30 / market total. Unverified. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://www.sensetime.com/en","title":"SenseTime site (SenseCore AI Infrastructure)","publisher":"SenseTime","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"No own-chip rental outside China is reported"}},"china":{"low":1.5,"central":1.5,"high":1.5,"unit":"percent","basis":"Share of AI-cloud market revenue in mainland China, calendar year 2025 (Omdia; includes model-as-a-service and platform revenue, wider than rented computing)","offBasis":true,"period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://eu.36kr.com/en/p/3924679520647302","title":"B-End Market Saturation Reached: Untapped C-End Frontier Becomes New Competitive Battlefield for Cloud Vendors to Seize Market Territory (citing Omdia, China AI cloud market 2025)","publisher":"36Kr","publishedDate":"2026-08-04","accessedDate":"2026-10-22","tier":"T5","locator":"Top five hold more than 80% of the RMB56.7B China AI cloud market (Omdia, via 36Kr); this provider is outside the top five"},"note":"5LayerCake T5 allocation: Omdia's top five hold more than 80% of the RMB56.7B China AI cloud market (36Kr, citing Omdia), leaving 19.9% for all others including Huawei Cloud (8%). The remaining 11.9% is shared among the other registered providers (China Mobile 3.0, China Unicom 1.8, Kingsoft Cloud 1.5, SenseTime 1.5, JD Cloud 1.0, UCloud 0.5, Capital Online 0.2) by size judgment, not by company data. Omdia's China AI cloud market for 2025 was RMB56.7B with Alibaba Cloud 38.1%, Volcano Engine 20.4%, Baidu 9.4%, Tencent Cloud 6.3% and eSurfing Cloud (China Telecom) 5.9%; the top five hold more than 80% (36Kr, 2026-08-04, citing Omdia; T4). A search summary of a Deep News article (not opened) says Omdia's measure covers IaaS, PaaS and model-as-a-service, which is wider than the card's rented-computing basis, so every China share is flagged offBasis. Sum of the five named = 80.1%."}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2},"china":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"c58759a83efe9cefa877794bd286717b70aa15bbc62faf8e91534aaf18683d4d"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"c58759a83efe9cefa877794bd286717b70aa15bbc62faf8e91534aaf18683d4d"}}},"nvidia":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"nvidia","name":"NVIDIA (DGX Cloud)","kind":"unit","commercialActivityVerified":false,"shares":{"gpu-capacity":{"low":0.058,"central":0.232,"high":0.929,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"FY2026 (to 25 Jan 2026); revenue estimated","periodEnd":"2026-01-25","source":{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","title":"NVIDIA Form 10-K for fiscal year ended January 25, 2026","publisher":"NVIDIA Corporation","publishedDate":"2026-02-25","accessedDate":"2026-10-22","tier":"T5","locator":"10-K: multi-year cloud service agreements ($27B at January 2026) are expected 'to support our research and development efforts'; risk factors: NVIDIA 'may not achieve any significant revenue from these offerings'"},"note":"5LayerCake T5 estimate. NVIDIA does not report DGX Cloud revenue; its cloud-service agreements are purchases from cloud providers for research and development. Assumed rental revenue of $0.2B. 200 / market total. Very weak evidence. No size indicator was found for this company, so the range runs from one quarter to four times the central estimate (the T5 default of -60% to +60% would be too narrow for a guess), which keeps it from being set aside on weak evidence. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","title":"NVIDIA Form 10-K for fiscal year ended January 25, 2026","publisher":"NVIDIA Corporation","publishedDate":"2026-02-25","accessedDate":"2026-10-22","tier":"T2","locator":"DGX Cloud runs on NVIDIA GPUs hosted by cloud providers; no cloud-owned custom chip"}},"china":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm","title":"NVIDIA Form 10-K for fiscal year ended January 25, 2026","publisher":"NVIDIA Corporation","publishedDate":"2026-02-25","accessedDate":"2026-10-22","tier":"T2","locator":"10-K: export controls could limit NVIDIA's 'ability to provide services such as NVIDIA AI cloud services'; no China rental is stated"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"601870a8156019271f439006b82c7c751a486a56f5948e443140ff1f1247878f"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"601870a8156019271f439006b82c7c751a486a56f5948e443140ff1f1247878f"}}},"rumble":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"rumble","name":"Rumble (Northern Data, Taiga Cloud)","kind":"public","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":0.232,"central":0.232,"high":0.232,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"FY2026 outlook (EUR 170-190M)","periodEnd":"2026-06-22","source":{"url":"https://pulse2.com/rumble-closes-acquisition-of-northern-data/","title":"Rumble closes acquisition of Northern Data","publisher":"Pulse 2.0","publishedDate":"2026-06-22","accessedDate":"2026-10-22","tier":"T5","locator":"'Roughly 22,000' H100/H200 GPUs, about 85% utilised in March 2026; Northern Data FY2026 revenue outlook raised to EUR 170-190 million; Rumble owns about 85.2%"},"note":"Company outlook as reported by Pulse 2.0: Northern Data FY2026 revenue EUR 170-190M, i.e. about $200M at an assumed 1.15 USD per EUR (rate not sourced). Includes some colocation, so rental may be lower. 22,000 GPUs x $11k x 85% = $0.2B agrees. 200 / market total. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://pulse2.com/rumble-closes-acquisition-of-northern-data/","title":"Rumble closes acquisition of Northern Data","publisher":"Pulse 2.0","publishedDate":"2026-06-22","accessedDate":"2026-10-22","tier":"T3","locator":"Article: H100 and H200 GPUs (NVIDIA); no Rumble-designed accelerator"}},"china":{"participates":false,"evidence":{"url":"https://pulse2.com/rumble-closes-acquisition-of-northern-data/","title":"Rumble closes acquisition of Northern Data","publisher":"Pulse 2.0","publishedDate":"2026-06-22","accessedDate":"2026-10-22","tier":"T3","locator":"Article: Northern Data/Rumble operations are in Europe and North America; no China"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"c5b7bbe3b1508f12b45f68fe38ee0a35f5f0c679797a9f70850140f2704f6cb6"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"c5b7bbe3b1508f12b45f68fe38ee0a35f5f0c679797a9f70850140f2704f6cb6"}}},"sakura-internet":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"sakura-internet","name":"Sakura Internet","kind":"public","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":0.135,"central":0.135,"high":0.135,"period":"Q1 FY3/2027 (April-June 2026), annualised","periodEnd":"2026-06-30","source":{"url":"https://www.sakura.ad.jp/corporate/wp-content/uploads/2026/07/en-260728-ir_1.pdf","title":"Summary of Financial Results for the First Quarter Ended June 30, 2026 (Consolidated) [JGAAP]","publisher":"SAKURA internet Inc.","publishedDate":"2026-07-28","accessedDate":"2026-10-22","tier":"T5","locator":"(ii) GPU infrastructure services: 'Net sales of GPU infrastructure services were 4,718,088 thousand yen (up 245.9% YoY)' from NVIDIA H200/B200 and H100 GPUs"},"note":"5LayerCake T5 share: the company's figure is a T1 results summary and the denominator is a 5LayerCake estimate. Sakura Internet's net sales of GPU infrastructure services, the line for its rented GPU services, were 4,718,088 thousand yen in the quarter ended 30 June 2026; x 4 = 18872 million yen; at 162.61 yen per US$ (Federal Reserve H.10, 30 June 2026, https://www.federalreserve.gov/releases/h10/hist/dat00_ja.htm) = $116.1M. 116.1 / 86,152. The summary does not split the line between rental to customers and any other service. Operations are in Japan (Ishikari Data Center). Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate.","unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars"},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://www.sakura.ad.jp/corporate/wp-content/uploads/2026/07/en-260728-ir_1.pdf","title":"Summary of Financial Results for the First Quarter Ended June 30, 2026 (Consolidated) [JGAAP]","publisher":"SAKURA internet Inc.","publishedDate":"2026-07-28","accessedDate":"2026-10-22","tier":"T1","locator":"GPU services use NVIDIA H100, H200 and B200 GPUs; no Sakura-designed accelerator mentioned"}},"china":{"participates":false,"evidence":{"url":"https://www.sakura.ad.jp/corporate/wp-content/uploads/2026/07/en-260728-ir_2.pdf","title":"Consolidated Financial Results for Q1 FY 3/2027 (presentation)","publisher":"SAKURA internet Inc.","publishedDate":"2026-07-28","accessedDate":"2026-10-22","tier":"T2","locator":"Presentation describes the Ishikari Data Center as 'the foundation of domestic AI infrastructure' supporting 'data sovereignty' and refers to 'Japanese GPU infrastructure'; no China business mentioned"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"c367a16a31646ac96fc6a4792af3fad18c3f90c236c1b58add7024dbc32a86dc"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"c367a16a31646ac96fc6a4792af3fad18c3f90c236c1b58add7024dbc32a86dc"}}},"mistral":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"mistral","name":"Mistral AI (Mistral Compute)","kind":"private","commercialActivityVerified":false,"shares":{"gpu-capacity":{"low":0.029,"central":0.116,"high":0.464,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"2026; revenue estimated","periodEnd":"2026-08-11","source":{"url":"https://mistral.ai/news/regional-inference-open-models-new-compute/","title":"In-region inference, open models, and new European infrastructure for sovereign AI","publisher":"Mistral AI","publishedDate":"2026-08-11","accessedDate":"2026-10-22","tier":"T5","locator":"Mistral Compute: anchor enterprises make multi-year commitments for European Compute Units; 'plans to build up to 1 GW of capacity by 2030'; no interim capacity"},"note":"5LayerCake T5 estimate. The August 2026 post gives no operating capacity or Mistral Compute revenue (Mistral's total run rate is above $400M per FT/PYMNTS, mostly models). Assumed $0.1B. 100 / market total. Very weak evidence. No size indicator was found for this company, so the range runs from one quarter to four times the central estimate (the T5 default of -60% to +60% would be too narrow for a guess), which keeps it from being set aside on weak evidence. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://mistral.ai/news/regional-inference-open-models-new-compute/","title":"In-region inference, open models, and new European infrastructure for sovereign AI","publisher":"Mistral AI","publishedDate":"2026-08-11","accessedDate":"2026-10-22","tier":"T2","locator":"Post describes Mistral-built infrastructure on NVIDIA systems; no Mistral-designed accelerator"}},"china":{"participates":false,"evidence":{"url":"https://mistral.ai/news/regional-inference-open-models-new-compute/","title":"In-region inference, open models, and new European infrastructure for sovereign AI","publisher":"Mistral AI","publishedDate":"2026-08-11","accessedDate":"2026-10-22","tier":"T2","locator":"Post is about European sovereign infrastructure; no China"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"faf0296466b2614a25f495aa22c9960deb40b4802c7cc722fdc5d87b2993458f"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"faf0296466b2614a25f495aa22c9960deb40b4802c7cc722fdc5d87b2993458f"}}},"naver-cloud":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"naver-cloud","name":"NAVER Cloud","kind":"unit","parentId":"naver","commercialActivityVerified":false,"shares":{"gpu-capacity":{"low":0.029,"central":0.116,"high":0.464,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"2026 (page undated); revenue estimated","periodEnd":"2026-10-22","source":{"url":"https://www.ncloud.com/","title":"NAVER Cloud Platform home page","publisher":"NAVER Cloud","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"GPU Server product under Compute; Seoul National University GPU case; no GPU count"},"note":"5LayerCake T5 estimate; the page gives no fleet or revenue. Assumed about 9,000 GPUs x $11k = $0.1B. 100 / market total. Very weak evidence. No size indicator was found for this company, so the range runs from one quarter to four times the central estimate (the T5 default of -60% to +60% would be too narrow for a guess), which keeps it from being set aside on weak evidence. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://www.ncloud.com/","title":"NAVER Cloud Platform home page","publisher":"NAVER Cloud","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Page lists a GPU Server product; no NAVER-designed accelerator mentioned"}},"china":{"participates":false,"evidence":{"url":"https://www.ncloud.com/","title":"NAVER Cloud Platform home page","publisher":"NAVER Cloud","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Korean platform; page does not mention China"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"6e69e53c35c907cb5a1801539401dc26935d8fa35b1f298f92f50a4580b39ca4"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"6e69e53c35c907cb5a1801539401dc26935d8fa35b1f298f92f50a4580b39ca4"}}},"tensorwave":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"tensorwave","name":"TensorWave","kind":"private","commercialActivityVerified":false,"shares":{"gpu-capacity":{"low":0.029,"central":0.116,"high":0.464,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"2026 (page undated); revenue estimated","periodEnd":"2026-10-22","source":{"url":"https://www.tensorwave.com/","title":"TensorWave home page","publisher":"TensorWave","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"AMD Instinct MI455X, MI355X, MI325X, MI300X clouds; customers shown include Meta, AMD, Moreh, Luma"},"note":"5LayerCake T5 estimate; the page gives no GPU count or revenue. Assumed about 8,000 GPUs x $11k x 100% = $0.1B. 100 / market total. Very weak evidence. No size indicator was found for this company, so the range runs from one quarter to four times the central estimate (the T5 default of -60% to +60% would be too narrow for a guess), which keeps it from being set aside on weak evidence. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://www.tensorwave.com/","title":"TensorWave home page","publisher":"TensorWave","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Page lists AMD Instinct merchant accelerators only"}},"china":{"participates":false,"evidence":{"url":"https://www.tensorwave.com/","title":"TensorWave home page","publisher":"TensorWave","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Page lists no locations and does not mention China"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"6126c5e6a0579ca8be7d7fd6fa846c3991720f03cd4297bab8403fc24ff2e9c4"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"6126c5e6a0579ca8be7d7fd6fa846c3991720f03cd4297bab8403fc24ff2e9c4"}}},"vultr":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"vultr","name":"Vultr","kind":"private","commercialActivityVerified":false,"shares":{"gpu-capacity":{"low":0.029,"central":0.116,"high":0.464,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"Valuation December 2024; revenue estimated","periodEnd":"2026-06-30","source":{"url":"https://www.alternativeswatch.com/2024/12/18/luminarx-capital-leads-growth-investment-in-cloud-provider-vultr-at-3-5bn-valuation/","title":"LuminArx Capital leads growth investment in cloud provider Vultr at $3.5bn valuation","publisher":"Alternatives Watch","publishedDate":"2024-12-18","accessedDate":"2026-10-22","tier":"T5","locator":"LuminArx and AMD Ventures invest $333M at a $3.5B valuation; 32 cloud data centres; AMD Instinct GPUs deployed"},"note":"5LayerCake T5 estimate with no revenue disclosed (Vultr's own site returned 403 and was not retried). Total revenue assumed about $0.3B (valuation $3.5B at about 10x), of which GPU rental about a third = $0.1B. 100 / market total. Very weak evidence. No size indicator was found for this company, so the range runs from one quarter to four times the central estimate (the T5 default of -60% to +60% would be too narrow for a guess), which keeps it from being set aside on weak evidence. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://www.alternativeswatch.com/2024/12/18/luminarx-capital-leads-growth-investment-in-cloud-provider-vultr-at-3-5bn-valuation/","title":"LuminArx Capital leads growth investment in cloud provider Vultr at $3.5bn valuation","publisher":"Alternatives Watch","publishedDate":"2024-12-18","accessedDate":"2026-10-22","tier":"T3","locator":"Article: AMD says Vultr deployed AMD Instinct accelerators; no Vultr-designed accelerator"}},"china":{"participates":false,"evidence":{"url":"https://www.alternativeswatch.com/2024/12/18/luminarx-capital-leads-growth-investment-in-cloud-provider-vultr-at-3-5bn-valuation/","title":"LuminArx Capital leads growth investment in cloud provider Vultr at $3.5bn valuation","publisher":"Alternatives Watch","publishedDate":"2024-12-18","accessedDate":"2026-10-22","tier":"T3","locator":"Article: 32 cloud data centres across six continents; none in mainland China is stated"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"9244d254e500758887d87a361779fddfed303c65cc0a179cfad742b58ce32187"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"9244d254e500758887d87a361779fddfed303c65cc0a179cfad742b58ce32187"}}},"gmo-internet":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"gmo-internet","name":"GMO Internet","kind":"public","commercialActivityVerified":false,"shares":{"gpu-capacity":{"low":0.023,"central":0.093,"high":0.371,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"2026; revenue estimated","periodEnd":"2026-10-22","source":{"url":"https://internet.gmo/en/","title":"GMO Internet Group home page","publisher":"GMO Internet","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Home page shows a 'GPU Cloud' banner (Japan-only)"},"note":"5LayerCake T5 estimate; no GPU count or revenue found. Assumed about $0.08B. 80 / market total. Unverified. No size indicator was found for this company, so the range runs from one quarter to four times the central estimate (the T5 default of -60% to +60% would be too narrow for a guess), which keeps it from being set aside on weak evidence. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://internet.gmo/en/","title":"GMO Internet Group home page","publisher":"GMO Internet","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Home page only; no GMO-designed accelerator mentioned"}},"china":{"participates":false,"evidence":{"url":"https://internet.gmo/en/","title":"GMO Internet Group home page","publisher":"GMO Internet","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Japanese service; page does not mention China"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"11c77c23900332c51361042c1d14e4bfa1cdfd49c189e61ee0c07812bcbeb54f"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"11c77c23900332c51361042c1d14e4bfa1cdfd49c189e61ee0c07812bcbeb54f"}}},"yotta":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"yotta","name":"Yotta Data Services (Shakti Cloud)","kind":"private","commercialActivityVerified":false,"shares":{"gpu-capacity":{"low":0.023,"central":0.093,"high":0.371,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"2026 (page undated); revenue estimated","periodEnd":"2026-10-22","source":{"url":"https://www.yotta.com/","title":"Yotta home page (Shakti Cloud)","publisher":"Yotta Data Services","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Shakti Cloud with NVIDIA H100 and L40S; Tier IV data centres including NM1 Mumbai; no GPU count"},"note":"5LayerCake T5 estimate; the page gives no GPU count or revenue. Assumed about 7,000 GPUs x $11k = $0.08B. 80 / market total. Very weak evidence. No size indicator was found for this company, so the range runs from one quarter to four times the central estimate (the T5 default of -60% to +60% would be too narrow for a guess), which keeps it from being set aside on weak evidence. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://www.yotta.com/","title":"Yotta home page (Shakti Cloud)","publisher":"Yotta Data Services","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Page lists NVIDIA GPUs only"}},"china":{"participates":false,"evidence":{"url":"https://www.yotta.com/","title":"Yotta home page (Shakti Cloud)","publisher":"Yotta Data Services","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Page names Mumbai (NM1) and India data centres; no China"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"10008041d233112297cfa8f3364ff9a66b981a47109984d89e4c29fd53e69883"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"10008041d233112297cfa8f3364ff9a66b981a47109984d89e4c29fd53e69883"}}},"nexgen-cloud":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"nexgen-cloud","name":"NexGen Cloud (Hyperstack)","kind":"private","commercialActivityVerified":true,"shares":{"gpu-capacity":{"low":0.081,"central":0.081,"high":0.488,"unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars","period":"CY2024","periodEnd":"2024-12-31","source":{"url":"https://edgeir.com/nexgen-cloud-raises-45m-to-build-europes-sovereign-ai-infrastructure-20250415","title":"NexGen Cloud raises $45M to build Europe's sovereign AI infrastructure","publisher":"EdgeIR","publishedDate":"2025-04-15","accessedDate":"2026-10-22","tier":"T5","locator":"'revenue rose from $3 million in 2023 to over $70m in 2024'; AI cloud operations revenue up 380%"},"note":"Press report of a company figure for CY2024 (over $70M; revenue was $3M in 2023, so the company grew about 23x in 2024); stale for a June 2026 annualised basis, so the engine also widens it. 2026 revenue is probably higher, so the high end runs to six times the 2024 figure ($420M). 70 / market total. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate."},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://edgeir.com/nexgen-cloud-raises-45m-to-build-europes-sovereign-ai-infrastructure-20250415","title":"NexGen Cloud raises $45M to build Europe's sovereign AI infrastructure","publisher":"EdgeIR","publishedDate":"2025-04-15","accessedDate":"2026-10-22","tier":"T3","locator":"Article: European sovereign AI cloud on NVIDIA GPUs; no NexGen-designed accelerator"}},"china":{"participates":false,"evidence":{"url":"https://edgeir.com/nexgen-cloud-raises-45m-to-build-europes-sovereign-ai-infrastructure-20250415","title":"NexGen Cloud raises $45M to build Europe's sovereign AI infrastructure","publisher":"EdgeIR","publishedDate":"2025-04-15","accessedDate":"2026-10-22","tier":"T3","locator":"Article: Europe-focused infrastructure; no China"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"3e4f52efa859f1fa98e8432effa9eaf39c1be27834b10257dad59c70d48c31f7"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"3e4f52efa859f1fa98e8432effa9eaf39c1be27834b10257dad59c70d48c31f7"}}},"softbank-corp":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"softbank-corp","name":"SoftBank Corp.","kind":"public","parentId":"softbank-group","commercialActivityVerified":false,"shares":{"gpu-capacity":{"low":0.013,"central":0.065,"high":0.427,"period":"2026; revenue estimated","periodEnd":"2026-10-22","source":{"url":"https://www.softbank.jp/en/corp/news/press/sbkk/2025/20251225_01","title":"SoftBank Launches AI Computing Platform Featuring Liquid-Cooled NVIDIA GB200 NVL72","publisher":"SoftBank Corp.","publishedDate":"2025-12-25","accessedDate":"2026-10-22","tier":"T5","locator":"'SoftBank's newly deployed platform comprises 1,224 NVIDIA Blackwell GPUs', 'designed for future expansion to more than 4,000 GPUs'; 'dedicated accelerated computing resource services'; plans to 'provide the infrastructure as a service to companies and research institutions in Japan' and use among 'its own group companies'; no revenue or price is given"},"note":"5LayerCake T5 estimate (assumed): the release shows a GPU rental service being launched but no customer, revenue or utilisation, so commercial activity is left unverified. Arithmetic: 1,224 GPUs x 8,760 hours x $10.50 an hour (CoreWeave's published on-demand list prices (https://www.coreweave.com/pricing, opened 2026-10-22): GB200 NVL72 $10.50 and HGX B200 $8.60 per GPU-hour) = $112.6M a year if fully rented at list. Central = half of that is rented to outside customers (the rest is used by group companies or unsold; 5LayerCake judgement) = $56.3M; low = one tenth = $11.3M; high = the 4,000 GPUs the platform is designed to reach, all rented at list = $367.9M. SoftBank's separate DGX B200 cluster of more than 4,000 GPUs (https://www.softbank.jp/en/corp/news/press/sbkk/2025/20250723_01, 2025-07-23) 'will initially be utilized by SB Intuitions' (a SoftBank subsidiary), so its own use is not counted. 56.3 / 86,152. The range is the researcher's own judgement, wider than the tier default. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate.","unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars"},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://www.softbank.jp/en/corp/news/press/sbkk/2025/20251225_01","title":"SoftBank Launches AI Computing Platform Featuring Liquid-Cooled NVIDIA GB200 NVL72","publisher":"SoftBank Corp.","publishedDate":"2025-12-25","accessedDate":"2026-10-22","tier":"T2","locator":"Platform is built on NVIDIA GB200 NVL72; no SoftBank-designed accelerator mentioned"}},"china":{"participates":false,"evidence":{"url":"https://www.softbank.jp/en/corp/news/press/sbkk/2025/20251225_01","title":"SoftBank Launches AI Computing Platform Featuring Liquid-Cooled NVIDIA GB200 NVL72","publisher":"SoftBank Corp.","publishedDate":"2025-12-25","accessedDate":"2026-10-22","tier":"T2","locator":"Service is for 'companies and research institutions in Japan' and SoftBank group companies; no China compute rental mentioned"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"0187e072903078fd3344efd0046ad77d14810bc6209badca21bf795dfc6d6ea3"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"0187e072903078fd3344efd0046ad77d14810bc6209badca21bf795dfc6d6ea3"}}},"kddi":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"kddi","name":"KDDI (GPU cloud)","kind":"unit","commercialActivityVerified":false,"shares":{"gpu-capacity":{"low":0.011,"central":0.053,"high":0.427,"period":"2026; revenue estimated","periodEnd":"2026-10-22","source":{"url":"https://newsroom.kddi.com/english/news/detail/kddi_nr-630_3986.html","title":"KDDI and HPE Join Forces to Launch AI Data Center Operations by Early 2026","publisher":"KDDI Corporation","publishedDate":"2025-06-26","accessedDate":"2026-10-22","tier":"T5","locator":"KDDI and HPE are 'collaborating to open the Osaka Sakai Data Center by early 2026' with 'a rack-scale system featuring the NVIDIA GB200 NVL72 platform'; KDDI plans to 'offer cloud-based AI computing services to organizations through WAKONX'; the release gives no GPU count, revenue or price"},"note":"5LayerCake T5 estimate (assumed): the release is a plan, not a service in operation, and no later page opened shows customers or revenue, so commercial activity is left unverified. The GPU count is not disclosed; assumed fleet 1,000 GB200 GPUs (a guess, about 14 NVL72 racks). Arithmetic: 1,000 x 8,760 hours x $10.50 an hour (CoreWeave's published on-demand list prices (https://www.coreweave.com/pricing, opened 2026-10-22): GB200 NVL72 $10.50 and HGX B200 $8.60 per GPU-hour) = $92.0M a year if fully rented at list. Central = half rented to outside customers (5LayerCake judgement) = $46.0M; low = one tenth = $9.2M; high = 4,000 GPUs all rented at list = $367.9M. 46.0 / 86,152. The range is the researcher's own judgement, wider than the tier default. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate.","unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars"},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://newsroom.kddi.com/english/news/detail/kddi_nr-630_3986.html","title":"KDDI and HPE Join Forces to Launch AI Data Center Operations by Early 2026","publisher":"KDDI Corporation","publishedDate":"2025-06-26","accessedDate":"2026-10-22","tier":"T2","locator":"Release names the NVIDIA GB200 NVL72 platform built by HPE; no KDDI-designed accelerator mentioned"}},"china":{"participates":false,"evidence":{"url":"https://newsroom.kddi.com/english/news/detail/kddi_nr-630_3986.html","title":"KDDI and HPE Join Forces to Launch AI Data Center Operations by Early 2026","publisher":"KDDI Corporation","publishedDate":"2025-06-26","accessedDate":"2026-10-22","tier":"T2","locator":"AI data center is in Sakai City, Osaka Prefecture, Japan; no China compute rental mentioned"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"f16929774525120fbf4247f8aaa366eb0265042bb0fa5728d33d4f88b29166eb"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"f16929774525120fbf4247f8aaa366eb0265042bb0fa5728d33d4f88b29166eb"}}},"ntt":{"schema":"5lc-index-inputs/1","key":"infrastructure-cloud","entityId":"ntt","name":"NTT (GPU cloud services)","kind":"unit","commercialActivityVerified":false,"shares":{"gpu-capacity":{"low":0.009,"central":0.044,"high":0.35,"period":"2026; revenue estimated","periodEnd":"2026-10-22","source":{"url":"https://www.nttdata.com/global/ja/news/topics/2025/031900/","title":"NVIDIA最新GPU を活用した大規模機械学習向け基盤提供を開始","publisher":"NTT DATA Corporation","publishedDate":"2025-03-19","accessedDate":"2026-10-22","tier":"T5","locator":"NTT DATA will start a 'GPU as a Service' in October 2025: NVIDIA DGX B200 systems (8 Blackwell GPUs each) in a DGX SuperPOD configuration, provided in a private environment for each customer, with GPU resources of top-class computing power in Japan (translated from the Japanese); the release gives no system count, revenue or price"},"note":"5LayerCake T5 estimate (assumed): the release announces a GPU rental service from October 2025 but no customers, revenue or GPU count, and no later page opened shows them, so commercial activity is left unverified. Assumed fleet 1,000 B200 GPUs (a guess, about 125 DGX B200 systems). Arithmetic: 1,000 x 8,760 hours x $8.60 an hour (CoreWeave's published on-demand list prices (https://www.coreweave.com/pricing, opened 2026-10-22): GB200 NVL72 $10.50 and HGX B200 $8.60 per GPU-hour) = $75.3M a year if fully rented at list. Central = half rented (5LayerCake judgement) = $37.7M; low = one tenth = $7.5M; high = 4,000 GPUs all rented at list = $301.3M. 37.7 / 86,152. The range is the researcher's own judgement, wider than the tier default. Market total about $86B a year for GPU computing outside China (5LayerCake estimate, Q2 2026 annualised: sum of provider estimates, with AWS 11.0, Azure 18.0, Google 9.0, Oracle 20.0, CoreWeave 10.3, Nebius 2.3 in $B, all other registered providers and about $6B for providers outside the register; full derivation and cross-checks in the aws.json note). Screening-grade T5 estimate.","unit":"percent","basis":"Share of revenue from AI compute rented to customers in the function, latest annualised, US dollars"},"custom-accelerator-capacity":{"participates":false,"evidence":{"url":"https://www.nttdata.com/global/ja/news/topics/2025/031900/","title":"NVIDIA最新GPU を活用した大規模機械学習向け基盤提供を開始","publisher":"NTT DATA Corporation","publishedDate":"2025-03-19","accessedDate":"2026-10-22","tier":"T2","locator":"Release describes only NVIDIA DGX B200 hardware; no NTT-designed accelerator offered for rent"}},"china":{"participates":false,"evidence":{"url":"https://www.nttdata.com/global/ja/news/topics/2025/031900/","title":"NVIDIA最新GPU を活用した大規模機械学習向け基盤提供を開始","publisher":"NTT DATA Corporation","publishedDate":"2025-03-19","accessedDate":"2026-10-22","tier":"T2","locator":"Service provides GPU resources in Japan for enterprises and research institutions; China appears only in a footnote about a ranking ('Excluding Chinese operators'), not as a market served"}}},"screening":{"gpu-capacity":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"65ca732c07b82b9bea13c900d368e629bdaae6bfc18597c0967cae9f6d26bbac"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"65ca732c07b82b9bea13c900d368e629bdaae6bfc18597c0967cae9f6d26bbac"}}}},"ratings":{"aws":{"a":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"aws","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"gpu-capacity":{"alternatives":{"level":2,"evidence":[{"url":"https://newsletter.semianalysis.com/p/clustermax-30-the-industry-standard","date":"2026-09-23","support":"ClusterMAX 3.0 summary: 'Nebius joins CoreWeave in the Platinum tier'; Google Cloud is placed in the Gold tier (full date shown)."},{"url":"https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/TranscriptFY26Q4.docx","date":"2026-07-29","support":"Microsoft says it added another gigawatt of capacity in the quarter and is on track to roughly double capacity in two years (FY26 Q4 call, full date)."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://newsletter.semianalysis.com/p/clustermax-30-the-industry-standard","date":"2026-09-23","support":"AWS section: 'GPU capacity must be secured in large blocks and far in advance' (full date shown)."},{"url":"https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/TranscriptFY26Q4.docx","date":"2026-07-29","support":"Microsoft CFO: 'Customer demand continues to exceed available capacity'; supply is short 'in a relatively extreme moment' (FY26 Q4 call transcript, full date)."},{"url":"https://epoch.ai/data-insights/data-centers-buildout-speeds","date":"2025-11-10","support":"Epoch AI: for sites started in the past three years, construction start to 1 GW took 1 to 3.6 years (full date shown)."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://investors.coreweave.com/news/news-details/2026/CoreWeave-Reports-Strong-Second-Quarter-2026-Results/","date":"2026-08-11","support":"CoreWeave says active power reached 1.5 GW after adding nearly 500 MW in the quarter; contracted power about 3.7 GW (Q2 2026 release, full date)."},{"url":"https://nextplatform.com/cloud/2026/09/14/despite-hefty-capital-investments-the-cloud-starts-paying-off-for-oracle/5296362","date":"2026-09-14","support":"Oracle added 850 MW of its own capacity in the quarter, and OpenAI has taken delivery of 618 MW at Abilene (The Next Platform, full date)."},{"url":"https://www.sec.gov/Archives/edgar/data/0001018724/000101872426000024/amzn-20260630xex991.htm","date":"2026-07-30","support":"The company says AWS's AI business passed a $25B run rate; it does not split out GPU rental (Q2 2026 release, full date)."}]}},"custom-accelerator-capacity":{"alternatives":{"level":4,"evidence":[{"url":"https://www.theregister.com/2026/04/29/amazon_chips_20b_business/","date":"2026-04-30","support":"Trainium is consumed through AWS; Jassy's mention of selling chips to third parties is hypothetical (The Register, full date)."},{"url":"https://www.anthropic.com/news/expanding-our-use-of-google-cloud-tpus-and-services","date":"2025-10-23","support":"Anthropic says it uses three chip platforms (Google TPUs, Amazon Trainium, NVIDIA GPUs) and expects well over 1 GW of TPU capacity in 2026 (full date)."},{"url":"https://www.thestandard.com.hk/tech-and-startup/article/325426/Meta-signs-multi-billion-dollar-deal-to-rent-Google-AI-chips-The-Information-reports/","date":"2026-02-27","support":"Reuters, citing The Information: Meta signed a multi-billion-dollar, multi-year deal to rent Google's TPUs (full date)."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://www.anthropic.com/news/anthropic-amazon-compute","date":"2026-04-20","support":"Anthropic says it uses over one million Trainium2 chips and expects nearly 1 GW of Trainium capacity by end-2026 (full date)."},{"url":"https://developers.googleblog.com/en/torchtpu-running-pytorch-natively-on-tpus-at-google-scale/","date":"2026-04-07","support":"Google says TorchTPU lets developers migrate PyTorch workloads to TPU with minimal code changes; porting tools are only now maturing (full date)."},{"url":"https://epoch.ai/data-insights/data-centers-buildout-speeds","date":"2025-11-10","support":"Epoch AI: for sites started in the past three years, construction start to 1 GW took 1 to 3.6 years (full date shown)."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://newsletter.semianalysis.com/p/aws-trainium3-deep-dive-a-potential","date":"2025-12-04","support":"SemiAnalysis credits TPUv7 with 'extremely strong perf per TCO', so TPU matches or exceeds Trainium on capability (full date)."},{"url":"https://www.futuriom.com/articles/news/google-raises-capex-wall-street-sells/2026/07","date":"2026-07-23","support":"Alphabet's CFO cites a supply-constrained environment: 'we plan to expand the use of third-party capacity' (Futuriom, full date)."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"aws","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"gpu-capacity":{"alternatives":{"level":2,"evidence":[{"url":"https://newsletter.semianalysis.com/i/217033016/nebius","date":"2026-09-23","support":"ClusterMAX 3.0 rated 77 GPU clouds (executive summary): Platinum CoreWeave and Nebius; Gold Google Cloud and Oracle; Silver Azure, Lambda, Firmus, TensorWave, GMI; Bronze Crusoe. Four large peers (Oracle, Azure, Google, CoreWeave) rate at or above AWS's Silver in 2.0."},{"url":"https://newsletter.semianalysis.com/p/clustermax-20-the-industry-standard","date":"2025-11-06","support":"ClusterMAX 2.0 (84 of 209 rated): AWS Silver; CoreWeave Platinum; Nebius, Oracle, Azure, Crusoe, Fluidstack Gold."},{"url":"https://storagenewsletter.com/2026/09/23/oracle-fiscal-1q27-financial-results","date":"2026-09-23","support":"Oracle says demand for AI cloud services keeps growing faster than supply; remaining performance obligations $664B; over 300,000 GPUs delivered since the end of Q4 FY26 (reported 23 Sep 2026). Shows another hyperscale supplier adding GPUs at AWS-like volume."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://datacenterknowledge.com/infrastructure/amazon-lifts-ai-infrastructure-spending-to-220b-as-demand-outpaces-capacity","date":"2026-07-31","support":"Jassy (Q2 call, as reported): AWS will lack enough capacity for 2026 demand, expects the constraint into 2027, and aims to double power capacity by end-2027; much 2027 capacity already reserved."},{"url":"https://epochai.substack.com/p/build-times-for-gigawatt-scale-data","date":"2025-11-10","support":"Epoch AI: AI data centres typically take 1 to 4 years from construction start to 1 GW; some hyperscalers plan 1-2 years, the fastest plan is about 12 months. Replacing a 12-13% share of rented GPU volume means new sites of this order."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.microsoft.com/en-us/investor/events/fy-2026/earnings-fy-2026-q4","date":"2026-07-29","support":"Microsoft's CFO says customer demand continues to exceed available capacity; the CEO says it added about a gigawatt in the quarter and will roughly double capacity in two years (call transcript, 29 Jul 2026). The nearest peers are also capacity constrained, so none could absorb AWS volume at once."},{"url":"https://storagenewsletter.com/2026/09/23/oracle-fiscal-1q27-financial-results","date":"2026-09-23","support":"Oracle says demand for AI cloud services keeps growing faster than supply; remaining performance obligations $664B; over 300,000 GPUs delivered since the end of Q4 FY26 (reported 23 Sep 2026)."}]}},"custom-accelerator-capacity":{"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 across Google TPUs, Amazon Trainium and NVIDIA GPUs and will use up to one million TPUs, so TPUs are a proven second platform."},{"url":"https://finance.yahoo.com/markets/stocks/article/google-to-sell-tpu-chips-to-select-customers-in-latest-shot-at-nvidia-214900221.html","date":"2026-04-29","support":"Alphabet will also sell TPUs for customers' own data centres; a multi-GW Anthropic TPU deal starts 2027. TPUs are the only other at-scale rented custom chip."},{"url":"https://blogs.microsoft.com/blog/2026/01/26/maia-200-the-ai-accelerator-built-for-inference/","date":"2026-01-26","support":"At launch (26 Jan 2026) Microsoft deployed Maia 200 in one region (Iowa, Arizona next), serving Microsoft and OpenAI models; customers reach it only indirectly through Foundry and Copilot."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://www.theregister.com/2026/04/10/amazon_may_sell_homegrown_chips/","date":"2026-04-10","support":"Trainium3 is 'nearly fully-subscribed'; a significant chunk of Trainium4, about 18 months from broad availability, is already reserved."},{"url":"https://datacenterknowledge.com/infrastructure/amazon-lifts-ai-infrastructure-spending-to-220b-as-demand-outpaces-capacity","date":"2026-07-31","support":"Anthropic and OpenAI have made multi-year, multi-gigawatt commitments to Trainium (as reported); moving that volume needs new capacity elsewhere."},{"url":"https://www.aboutamazon.com/news/aws/aws-project-rainier-ai-trainium-chips-compute-cluster","date":"undated","support":"Amazon says Project Rainier (nearly half a million Trainium2 chips) was built in under a year and Claude was expected on over 1 million Trainium2 chips; page shows no date."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://techcrunch.com/2026/04/29/google-cloud-surpasses-20b-but-says-growth-was-capacity-constrained/","date":"2026-04-29","support":"Pichai says Google is compute constrained near term and Cloud revenue would have been higher with more supply; backlog doubled to $462B (Q1 call, 29 Apr 2026). TPU capacity, the only comparable alternative, has no spare room."},{"url":"https://www.theregister.com/2026/04/10/amazon_may_sell_homegrown_chips/","date":"2026-04-10","support":"Trainium3 only shipped in early 2026 and is nearly fully subscribed; the alternative platform would need to roughly double to take this volume."}]}}}}},"google-cloud":{"a":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"google-cloud","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"gpu-capacity":{"alternatives":{"level":2,"evidence":[{"url":"https://newsletter.semianalysis.com/p/clustermax-30-the-industry-standard","date":"2026-09-23","support":"ClusterMAX 3.0 summary: 'Nebius joins CoreWeave in the Platinum tier'; Google Cloud is placed in the Gold tier (full date shown)."},{"url":"https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/TranscriptFY26Q4.docx","date":"2026-07-29","support":"Microsoft says it added another gigawatt of capacity in the quarter and is on track to roughly double capacity in two years (FY26 Q4 call, full date)."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://www.futuriom.com/articles/news/google-raises-capex-wall-street-sells/2026/07","date":"2026-07-23","support":"Alphabet's CFO cites a supply-constrained environment: 'we plan to expand the use of third-party capacity' (Futuriom, full date)."},{"url":"https://epoch.ai/data-insights/data-centers-buildout-speeds","date":"2025-11-10","support":"Epoch AI: for sites started in the past three years, construction start to 1 GW took 1 to 3.6 years (full date shown)."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://investors.coreweave.com/news/news-details/2026/CoreWeave-Reports-Strong-Second-Quarter-2026-Results/","date":"2026-08-11","support":"CoreWeave says active power reached 1.5 GW after adding nearly 500 MW in the quarter; contracted power about 3.7 GW (Q2 2026 release, full date)."},{"url":"https://www.sec.gov/Archives/edgar/data/0001513845/000110465926094568/tm2622968d1_ex99-2.htm","date":"2026-08-12","support":"Nebius expects 5 GW of contracted power by end-2026 and plans to deploy more than 1 GW a year (Q2 2026 letter, full date)."}]}},"custom-accelerator-capacity":{"alternatives":{"level":4,"evidence":[{"url":"https://www.theregister.com/2026/04/29/amazon_chips_20b_business/","date":"2026-04-30","support":"Trainium is consumed through AWS; Jassy's mention of selling chips to third parties is hypothetical (The Register, full date)."},{"url":"https://www.anthropic.com/news/expanding-our-use-of-google-cloud-tpus-and-services","date":"2025-10-23","support":"Anthropic says it uses three chip platforms (Google TPUs, Amazon Trainium, NVIDIA GPUs) and expects well over 1 GW of TPU capacity in 2026 (full date)."}]},"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 says it uses three chip platforms (Google TPUs, Amazon Trainium, NVIDIA GPUs) and expects well over 1 GW of TPU capacity in 2026 (full date)."},{"url":"https://www.thestandard.com.hk/tech-and-startup/article/325426/Meta-signs-multi-billion-dollar-deal-to-rent-Google-AI-chips-The-Information-reports/","date":"2026-02-27","support":"Reuters, citing The Information: Meta signed a multi-billion-dollar, multi-year deal to rent Google's TPUs (full date)."},{"url":"https://epoch.ai/data-insights/data-centers-buildout-speeds","date":"2025-11-10","support":"Epoch AI: for sites started in the past three years, construction start to 1 GW took 1 to 3.6 years (full date shown)."}]},"capabilityGap":{"level":4,"evidence":[{"url":"https://www.theregister.com/2026/04/29/amazon_chips_20b_business/","date":"2026-04-30","support":"Jassy: Trainium2 'has largely sold out', Trainium3 'is nearly fully subscribed', Trainium4 already reserved (The Register, full date)."},{"url":"https://newsletter.semianalysis.com/p/aws-trainium3-deep-dive-a-potential","date":"2025-12-04","support":"SemiAnalysis: Trainium3 lacks NVFP4 and AWS did not plan LNC=8 support until mid-2026, so its software trails TPU's (full date)."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"google-cloud","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"gpu-capacity":{"alternatives":{"level":2,"evidence":[{"url":"https://newsletter.semianalysis.com/i/217033016/nebius","date":"2026-09-23","support":"ClusterMAX 3.0 rated 77 GPU clouds (executive summary): Platinum CoreWeave and Nebius; Gold Google Cloud and Oracle; Silver Azure, Lambda, Firmus, TensorWave, GMI; Bronze Crusoe. Google is Gold; Oracle, Azure, AWS and CoreWeave are the other large GPU clouds."},{"url":"https://storagenewsletter.com/2026/09/23/oracle-fiscal-1q27-financial-results","date":"2026-09-23","support":"Oracle says demand for AI cloud services keeps growing faster than supply; remaining performance obligations $664B; over 300,000 GPUs delivered since the end of Q4 FY26 (reported 23 Sep 2026)."},{"url":"https://investors.coreweave.com/news/news-details/2026/CoreWeave-Reports-Strong-Second-Quarter-2026-Results/","date":"2026-08-11","support":"CoreWeave says Q2 2026 revenue was $2,575M, backlog about $104B, active power 1.5 GW and contracted power about 3.7 GW (company release, 11 Aug 2026)."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://techcrunch.com/2026/04/29/google-cloud-surpasses-20b-but-says-growth-was-capacity-constrained/","date":"2026-04-29","support":"Pichai says Google is compute constrained near term and Cloud revenue would have been higher with more supply; backlog doubled to $462B (Q1 call, 29 Apr 2026)."},{"url":"https://abc.xyz/investor/events/event-details/2026/2026-Q2-Earnings-Call-2026-GgTAq7Is0z/default.aspx","date":"2026-07-22","support":"Alphabet's CFO and CEO say Google remains supply constrained; backlog $514B; it will add third-party capacity in Q3 as a bridge (Q2 call, 22 Jul 2026)."},{"url":"https://epochai.substack.com/p/build-times-for-gigawatt-scale-data","date":"2025-11-10","support":"Epoch AI: AI data centres typically take 1 to 4 years from construction start to 1 GW; some hyperscalers plan 1-2 years, the fastest plan is about 12 months."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.microsoft.com/en-us/investor/events/fy-2026/earnings-fy-2026-q4","date":"2026-07-29","support":"Microsoft's CFO says customer demand continues to exceed available capacity; the CEO says it added about a gigawatt in the quarter and will roughly double capacity in two years (call transcript, 29 Jul 2026). Peers are also capacity constrained."},{"url":"https://datacenterknowledge.com/infrastructure/amazon-lifts-ai-infrastructure-spending-to-220b-as-demand-outpaces-capacity","date":"2026-07-31","support":"Jassy (Q2 call, as reported): AWS will lack enough capacity for 2026 demand, expects the constraint into 2027, and aims to double power capacity by end-2027; much 2027 capacity already reserved."}]}},"custom-accelerator-capacity":{"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 across Google TPUs, Amazon Trainium and NVIDIA GPUs, so Trainium and GPUs substitute for TPU workloads."},{"url":"https://datacenterknowledge.com/infrastructure/amazon-lifts-ai-infrastructure-spending-to-220b-as-demand-outpaces-capacity","date":"2026-07-31","support":"Anthropic and OpenAI have made multi-year, multi-gigawatt commitments to Trainium (as reported), the second at-scale rented custom chip."},{"url":"https://blogs.microsoft.com/blog/2026/01/26/maia-200-the-ai-accelerator-built-for-inference/","date":"2026-01-26","support":"At launch (26 Jan 2026) Microsoft deployed Maia 200 in one region, serving Microsoft and OpenAI models; customers reach it only indirectly through Foundry and Copilot."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://finance.yahoo.com/markets/stocks/article/google-to-sell-tpu-chips-to-select-customers-in-latest-shot-at-nvidia-214900221.html","date":"2026-04-29","support":"A multi-GW next-generation TPU agreement with Anthropic starts coming online in 2027; Alphabet is also selling TPUs for customers' own sites."},{"url":"https://techcrunch.com/2026/04/29/google-cloud-surpasses-20b-but-says-growth-was-capacity-constrained/","date":"2026-04-29","support":"Pichai says Google is compute constrained near term and Cloud revenue would have been higher with more supply; backlog doubled to $462B (Q1 call, 29 Apr 2026)."},{"url":"https://www.anthropic.com/news/expanding-our-use-of-google-cloud-tpus-and-services","date":"2025-10-23","support":"Anthropic says up to one million TPUs, well over a gigawatt online in 2026: volume that would need replacing."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://www.theregister.com/2026/04/10/amazon_may_sell_homegrown_chips/","date":"2026-04-10","support":"Trainium3 only shipped in early 2026 and is nearly fully subscribed; Trainium4 is about 18 months from broad availability."},{"url":"https://datacenterknowledge.com/infrastructure/amazon-lifts-ai-infrastructure-spending-to-220b-as-demand-outpaces-capacity","date":"2026-07-31","support":"AWS says it cannot meet 2026 demand and expects constraints into 2027, so Trainium cannot absorb TPU volume quickly."}]}}}}},"oracle-cloud-infrastructure":{"a":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"oracle-cloud-infrastructure","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"gpu-capacity":{"alternatives":{"level":2,"evidence":[{"url":"https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/TranscriptFY26Q4.docx","date":"2026-07-29","support":"Microsoft says it added another gigawatt of capacity in the quarter and is on track to roughly double capacity in two years (FY26 Q4 call, full date)."},{"url":"https://investors.coreweave.com/news/news-details/2026/CoreWeave-Reports-Strong-Second-Quarter-2026-Results/","date":"2026-08-11","support":"CoreWeave says active power reached 1.5 GW after adding nearly 500 MW in the quarter; contracted power about 3.7 GW (Q2 2026 release, full date)."},{"url":"https://newsletter.semianalysis.com/p/clustermax-30-the-industry-standard","date":"2026-09-23","support":"Several other large GPU clouds are rated (CoreWeave and Nebius Platinum, Google Cloud Gold), with Azure and AWS also at gigawatt scale (full date)."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://nextplatform.com/cloud/2026/09/14/despite-hefty-capital-investments-the-cloud-starts-paying-off-for-oracle/5296362","date":"2026-09-14","support":"The Next Platform reports Oracle GPU utilisation of 97.9% in Q1 FY27 and that GPU and XPU demand far outstripped supply (full date)."},{"url":"https://epoch.ai/data-insights/data-centers-buildout-speeds","date":"2025-11-10","support":"Epoch AI: for sites started in the past three years, construction start to 1 GW took 1 to 3.6 years (full date shown)."},{"url":"https://nextplatform.com/cloud/2026/09/14/despite-hefty-capital-investments-the-cloud-starts-paying-off-for-oracle/5296362","date":"2026-09-14","support":"About a quarter of the GPU rental market runs at 97.9% utilisation; replacing it needs new buildings, not spare racks (5LayerCake reading of The Next Platform)."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/TranscriptFY26Q4.docx","date":"2026-07-29","support":"Microsoft CFO: 'Customer demand continues to exceed available capacity'; supply is short 'in a relatively extreme moment' (FY26 Q4 call transcript, full date)."},{"url":"https://nextplatform.com/cloud/2026/09/14/despite-hefty-capital-investments-the-cloud-starts-paying-off-for-oracle/5296362","date":"2026-09-14","support":"Oracle added 850 MW of its own capacity in the quarter, and OpenAI has taken delivery of 618 MW at Abilene (The Next Platform, full date)."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"oracle-cloud-infrastructure","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"gpu-capacity":{"alternatives":{"level":2,"evidence":[{"url":"https://newsletter.semianalysis.com/i/217033016/nebius","date":"2026-09-23","support":"ClusterMAX 3.0 rated 77 GPU clouds (executive summary): Platinum CoreWeave and Nebius; Gold Google Cloud and Oracle; Silver Azure, Lambda, Firmus, TensorWave, GMI; Bronze Crusoe. Oracle is Gold; Azure, AWS, Google and CoreWeave are the other large GPU clouds."},{"url":"https://www.microsoft.com/en-us/investor/events/fy-2026/earnings-fy-2026-q4","date":"2026-07-29","support":"Microsoft's CFO says customer demand continues to exceed available capacity; the CEO says it added about a gigawatt in the quarter and will roughly double capacity in two years (call transcript, 29 Jul 2026)."},{"url":"https://investors.coreweave.com/news/news-details/2026/CoreWeave-Reports-Strong-Second-Quarter-2026-Results/","date":"2026-08-11","support":"CoreWeave says Q2 2026 revenue was $2,575M, backlog about $104B, active power 1.5 GW and contracted power about 3.7 GW (company release, 11 Aug 2026)."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://storagenewsletter.com/2026/09/23/oracle-fiscal-1q27-financial-results","date":"2026-09-23","support":"Oracle says demand for AI cloud services keeps growing faster than supply; remaining performance obligations $664B; over 300,000 GPUs delivered since the end of Q4 FY26 (reported 23 Sep 2026)."},{"url":"https://epochai.substack.com/p/build-times-for-gigawatt-scale-data","date":"2025-11-10","support":"Epoch AI: AI data centres typically take 1 to 4 years from construction start to 1 GW; some hyperscalers plan 1-2 years, the fastest plan is about 12 months. Oracle's ~23% share of rented GPU volume implies several gigawatts of replacement sites."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.microsoft.com/en-us/investor/events/fy-2026/earnings-fy-2026-q4","date":"2026-07-29","support":"Microsoft's CFO says customer demand continues to exceed available capacity; the CEO says it added about a gigawatt in the quarter and will roughly double capacity in two years (call transcript, 29 Jul 2026). Azure is the closest in scale and also supply constrained."},{"url":"https://datacenterknowledge.com/infrastructure/amazon-lifts-ai-infrastructure-spending-to-220b-as-demand-outpaces-capacity","date":"2026-07-31","support":"Jassy (Q2 call, as reported): AWS will lack enough capacity for 2026 demand, expects the constraint into 2027, and aims to double power capacity by end-2027; much 2027 capacity already reserved."}]}}}}},"microsoft-intelligent-cloud":{"a":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"microsoft-intelligent-cloud","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"gpu-capacity":{"alternatives":{"level":2,"evidence":[{"url":"https://newsletter.semianalysis.com/p/clustermax-30-the-industry-standard","date":"2026-09-23","support":"ClusterMAX 3.0 summary: 'Nebius joins CoreWeave in the Platinum tier'; Google Cloud is placed in the Gold tier (full date shown)."},{"url":"https://nextplatform.com/cloud/2026/09/14/despite-hefty-capital-investments-the-cloud-starts-paying-off-for-oracle/5296362","date":"2026-09-14","support":"Oracle added 850 MW of its own capacity in the quarter, and OpenAI has taken delivery of 618 MW at Abilene (The Next Platform, full date)."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/TranscriptFY26Q4.docx","date":"2026-07-29","support":"Microsoft says it added another gigawatt of capacity in the quarter and is on track to roughly double capacity in two years (FY26 Q4 call, full date)."},{"url":"https://epoch.ai/data-insights/data-centers-buildout-speeds","date":"2025-11-10","support":"Epoch AI: for sites started in the past three years, construction start to 1 GW took 1 to 3.6 years (full date shown)."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://nextplatform.com/cloud/2026/09/14/despite-hefty-capital-investments-the-cloud-starts-paying-off-for-oracle/5296362","date":"2026-09-14","support":"The Next Platform reports Oracle GPU utilisation of 97.9% in Q1 FY27 and that GPU and XPU demand far outstripped supply (full date)."},{"url":"https://www.sec.gov/Archives/edgar/data/0001513845/000110465926094568/tm2622968d1_ex99-2.htm","date":"2026-08-12","support":"Nebius says 'Capacity remains a binding constraint on AI adoption' and it could sell all its 2027 capacity today (Q2 2026 letter, full date)."},{"url":"https://www.futuriom.com/articles/news/google-raises-capex-wall-street-sells/2026/07","date":"2026-07-23","support":"Alphabet's CFO cites a supply-constrained environment: 'we plan to expand the use of third-party capacity' (Futuriom, full date)."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"microsoft-intelligent-cloud","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"gpu-capacity":{"alternatives":{"level":2,"evidence":[{"url":"https://newsletter.semianalysis.com/i/217033016/nebius","date":"2026-09-23","support":"ClusterMAX 3.0 rated 77 GPU clouds (executive summary): Platinum CoreWeave and Nebius; Gold Google Cloud and Oracle; Silver Azure, Lambda, Firmus, TensorWave, GMI; Bronze Crusoe. Four large peers (Oracle, AWS, Google, CoreWeave) offer comparable scale; Azure's own rating fell to Silver."},{"url":"https://storagenewsletter.com/2026/09/23/oracle-fiscal-1q27-financial-results","date":"2026-09-23","support":"Oracle says demand for AI cloud services keeps growing faster than supply; remaining performance obligations $664B; over 300,000 GPUs delivered since the end of Q4 FY26 (reported 23 Sep 2026)."},{"url":"https://investors.coreweave.com/news/news-details/2026/CoreWeave-Reports-Strong-Second-Quarter-2026-Results/","date":"2026-08-11","support":"CoreWeave says Q2 2026 revenue was $2,575M, backlog about $104B, active power 1.5 GW and contracted power about 3.7 GW (company release, 11 Aug 2026)."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://www.microsoft.com/en-us/investor/events/fy-2026/earnings-fy-2026-q4","date":"2026-07-29","support":"Microsoft's CFO says customer demand continues to exceed available capacity; the CEO says it added about a gigawatt in the quarter and will roughly double capacity in two years (call transcript, 29 Jul 2026)."},{"url":"https://epochai.substack.com/p/build-times-for-gigawatt-scale-data","date":"2025-11-10","support":"Epoch AI: AI data centres typically take 1 to 4 years from construction start to 1 GW; some hyperscalers plan 1-2 years, the fastest plan is about 12 months. Azure's ~21% share of rented GPU volume implies several gigawatts of new sites."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://datacenterknowledge.com/infrastructure/amazon-lifts-ai-infrastructure-spending-to-220b-as-demand-outpaces-capacity","date":"2026-07-31","support":"Jassy (Q2 call, as reported): AWS will lack enough capacity for 2026 demand, expects the constraint into 2027, and aims to double power capacity by end-2027; much 2027 capacity already reserved. Peers are short of capacity too, so the nearest alternatives cannot take Azure's volume at once."},{"url":"https://storagenewsletter.com/2026/09/23/oracle-fiscal-1q27-financial-results","date":"2026-09-23","support":"Oracle says demand for AI cloud services keeps growing faster than supply; remaining performance obligations $664B; over 300,000 GPUs delivered since the end of Q4 FY26 (reported 23 Sep 2026). Oracle's scale is comparable to Azure's."}]}}}}},"coreweave":{"a":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"coreweave","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"gpu-capacity":{"alternatives":{"level":2,"evidence":[{"url":"https://newsletter.semianalysis.com/p/clustermax-30-the-industry-standard","date":"2026-09-23","support":"ClusterMAX 3.0 summary: 'Nebius joins CoreWeave in the Platinum tier'; Google Cloud is placed in the Gold tier (full date shown)."},{"url":"https://nextplatform.com/cloud/2026/09/14/despite-hefty-capital-investments-the-cloud-starts-paying-off-for-oracle/5296362","date":"2026-09-14","support":"Oracle added 850 MW of its own capacity in the quarter, and OpenAI has taken delivery of 618 MW at Abilene (The Next Platform, full date)."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://investors.coreweave.com/news/news-details/2026/CoreWeave-Reports-Strong-Second-Quarter-2026-Results/","date":"2026-08-11","support":"CoreWeave reports revenue backlog of about $104B at 30 June 2026, excluding over $25B signed early in Q3 (release, full date)."},{"url":"https://epoch.ai/data-insights/data-centers-buildout-speeds","date":"2025-11-10","support":"Epoch AI: for sites started in the past three years, construction start to 1 GW took 1 to 3.6 years (full date shown)."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/0001513845/000110465926094568/tm2622968d1_ex99-2.htm","date":"2026-08-12","support":"Nebius expects 5 GW of contracted power by end-2026 and plans to deploy more than 1 GW a year (Q2 2026 letter, full date)."},{"url":"https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/TranscriptFY26Q4.docx","date":"2026-07-29","support":"Microsoft says it added another gigawatt of capacity in the quarter and is on track to roughly double capacity in two years (FY26 Q4 call, full date)."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"coreweave","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"gpu-capacity":{"alternatives":{"level":2,"evidence":[{"url":"https://newsletter.semianalysis.com/i/217033016/nebius","date":"2026-09-23","support":"ClusterMAX 3.0 rated 77 GPU clouds (executive summary): Platinum CoreWeave and Nebius; Gold Google Cloud and Oracle; Silver Azure, Lambda, Firmus, TensorWave, GMI; Bronze Crusoe. Only Nebius matches CoreWeave's Platinum; Oracle and Google (Gold) are at scale."},{"url":"https://newsletter.semianalysis.com/p/clustermax-20-the-industry-standard","date":"2025-11-06","support":"ClusterMAX 2.0: CoreWeave was the only Platinum provider; Nebius, Oracle, Azure, Crusoe and Fluidstack were Gold."},{"url":"https://storagenewsletter.com/2026/09/23/oracle-fiscal-1q27-financial-results","date":"2026-09-23","support":"Oracle says demand for AI cloud services keeps growing faster than supply; remaining performance obligations $664B; over 300,000 GPUs delivered since the end of Q4 FY26 (reported 23 Sep 2026)."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://investors.coreweave.com/news/news-details/2026/CoreWeave-Reports-Strong-Second-Quarter-2026-Results/","date":"2026-08-11","support":"CoreWeave says Q2 2026 revenue was $2,575M, backlog about $104B, active power 1.5 GW and contracted power about 3.7 GW (company release, 11 Aug 2026). Contracted minus active power leaves about 2.2 GW still to come online (our arithmetic)."},{"url":"https://epochai.substack.com/p/build-times-for-gigawatt-scale-data","date":"2025-11-10","support":"Epoch AI: AI data centres typically take 1 to 4 years from construction start to 1 GW; some hyperscalers plan 1-2 years, the fastest plan is about 12 months."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://assets.nebius.com/assets/4462517b-ce83-41f2-96ed-f2ac1bc06a05/SHLQ226%20(1).pdf-0","date":"2026-08-12","support":"Nebius reports annualised run-rate revenue of $3.0B at end-June 2026 (letter, 12 Aug 2026), under a third of CoreWeave's Q2 revenue run-rate ($2,575M x 4 = $10.3B, our arithmetic)."},{"url":"https://newsletter.semianalysis.com/i/217033016/nebius","date":"2026-09-23","support":"Gold-rated Oracle and Google are the large-scale alternatives; Nebius is the other Platinum."}]}}}}},"alibaba-cloud":{"a":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"alibaba-cloud","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"china":{"alternatives":{"level":3,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"36Kr, citing Omdia: 2025 China AI cloud shares Alibaba 38.1%, Volcano 20.4%, Baidu 9.4%, Tencent 6.3%, eSurfing 5.9%; top five over 80% (full date)."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://technode.com/2025/11/26/alibaba-ceo-says-ai-bubble-unlikely-in-next-three-years/","date":"2025-11-26","support":"Alibaba CEO Eddie Wu says AI resources will remain in short supply, with no bubble likely in three years (TechNode, full date)."},{"url":"https://www.businessday.co.za/world/international-companies/2026-03-18-tencent-to-ramp-up-ai-investment-after-chip-curbs-slowed-spending/","date":"2026-03-18","support":"Reuters: Tencent said chip export restrictions held back its 2025 spending; capex fell short of internal targets on limited access to AI chips (full date)."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/0001577552/000110465926099220/tm2623667d1_ex99-1.htm","date":"2026-08-20","support":"The company says its own T-Head Zhenwu chips have broad adoption via Alibaba Cloud, with more than 650 external customers (June quarter 2026 release, full date)."},{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"Omdia via 36Kr: the next largest provider, Volcano Engine, holds 20.4%, about half Alibaba Cloud's 38.1% (full date)."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"alibaba-cloud","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"china":{"alternatives":{"level":2,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"36Kr, citing Omdia data released 19 May: China's AI cloud market was 56.7B yuan in 2025; Alibaba 38.1%, Volcano Engine 20.4%, Baidu 9.4%, Tencent 6.3%, eSurfing (China Telecom) 5.9%; top five over 80%. Four peers each run AI clouds at scale, none above 21%."},{"url":"https://www.huaweicloud.com/eu/news/20250919133255709.html","date":"2025-09-19","support":"Huawei says users of its AI Compute Service (CloudMatrix384) rose from 321 in 2024 to 1,805 in 2025, a fifth credible domestic alternative (company announcement, 19 Sep 2025)."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://www.trendforce.com/news/?p=60480","date":"2026-07-09","support":"TrendForce (9 Jul 2026): Beijing may approve under 200,000 NVIDIA H200 chips, under half of what firms sought; Chinese AI developers are racing to build in-house chips to cope with shortages; inference is steered to domestic chips."},{"url":"https://newsletter.semianalysis.com/p/huawei-ascend-production-ramp","date":"2025-09-08","support":"SemiAnalysis (8 Sep 2025): about 805k Ascend units projected for 2025; HBM production is the bottleneck, and Huawei may not make 1M Ascend chips next year without more foreign HBM."},{"url":"https://www.businessday.co.za/world/international-companies/2026-08-20-alibaba-misses-profit-target-as-capex-climbs-75-in-ai-push/","date":"2026-08-20","support":"Reuters via BusinessDay (20 Aug 2026): Alibaba's June-quarter capex rose 75% to 67.68B yuan as it keeps building AI capacity."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"Alibaba's 38.1% share is almost twice Volcano Engine's 20.4%, the next largest (Omdia via 36Kr, 4 Aug 2026)."},{"url":"https://www.businessday.co.za/world/international-companies/2026-08-20-alibaba-misses-profit-target-as-capex-climbs-75-in-ai-push/","date":"2026-08-20","support":"Alibaba's AI cloud and compute revenue was 48.44B yuan in one quarter (Reuters via BusinessDay, 20 Aug 2026)."},{"url":"https://www.prnewswire.com/news-releases/baidu-announces-second-quarter-2026-results-302853860.html","date":"2026-08-18","support":"Baidu says its entire AI Cloud Infra revenue was RMB 7.3B in the same quarter, so mid-size peers fall well short of Alibaba's volume (company release, 18 Aug 2026)."}]}}}}},"volcano-engine":{"a":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"volcano-engine","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"china":{"alternatives":{"level":2,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"36Kr, citing Omdia: 2025 China AI cloud shares Alibaba 38.1%, Volcano 20.4%, Baidu 9.4%, Tencent 6.3%, eSurfing 5.9%; top five over 80% (full date)."},{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"36Kr: China cloud vendors' 'products and basic performance are gradually converging'; customers pick the cheapest when performance is close (full date)."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://kr-asia.com/bytedance-raises-volcano-engines-maas-revenue-target-on-seedance-2-0-growth","date":"2026-06-05","support":"KrASIA: Seedance users faced long queues; ByteDance used whitelists to give annual-contract customers stable access (full date)."},{"url":"https://www.businessday.co.za/world/international-companies/2026-03-18-tencent-to-ramp-up-ai-investment-after-chip-curbs-slowed-spending/","date":"2026-03-18","support":"Reuters: Tencent said chip export restrictions held back its 2025 spending; capex fell short of internal targets on limited access to AI chips (full date)."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"Omdia via 36Kr: only Alibaba Cloud (38.1%) is larger than Volcano Engine (20.4%); the rest are under 10% each (full date)."},{"url":"https://technode.com/2025/11/26/alibaba-ceo-says-ai-bubble-unlikely-in-next-three-years/","date":"2025-11-26","support":"Alibaba CEO Eddie Wu says AI resources will remain in short supply, with no bubble likely in three years (TechNode, full date)."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"volcano-engine","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"china":{"alternatives":{"level":2,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"36Kr, citing Omdia data released 19 May: China's AI cloud market was 56.7B yuan in 2025; Alibaba 38.1%, Volcano Engine 20.4%, Baidu 9.4%, Tencent 6.3%, eSurfing (China Telecom) 5.9%; top five over 80%. Alibaba, Baidu, Tencent and eSurfing are the other large AI clouds."},{"url":"https://www.huaweicloud.com/eu/news/20250919133255709.html","date":"2025-09-19","support":"Huawei says its AI Compute Service had 1,805 users in 2025, up from 321, on CloudMatrix384 (company announcement, 19 Sep 2025)."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://finance.yahoo.com/news/tiktok-parent-plans-14-billion-133112137.html","date":"2026-01-02","support":"Benzinga (2 Jan 2026), citing SCMP: ByteDance plans about 100B yuan on NVIDIA chips in 2026; Beijing had not formally approved H200 imports."},{"url":"https://www.trendforce.com/news/?p=60480","date":"2026-07-09","support":"TrendForce (9 Jul 2026): Beijing may approve under 200,000 NVIDIA H200 chips, under half of what firms sought; Chinese AI developers are racing to build in-house chips to cope with shortages; inference is steered to domestic chips."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"Alibaba (38.1%) is almost twice Volcano's size and fully capable, so the gap is small; no one has spare chips (Omdia via 36Kr, 4 Aug 2026)."},{"url":"https://www.trendforce.com/news/?p=60480","date":"2026-07-09","support":"TrendForce (9 Jul 2026): Beijing may approve under 200,000 NVIDIA H200 chips, under half of what firms sought; Chinese AI developers are racing to build in-house chips to cope with shortages; inference is steered to domestic chips."}]}}}}},"nebius":{"a":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"nebius","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"gpu-capacity":{"alternatives":{"level":2,"evidence":[{"url":"https://newsletter.semianalysis.com/p/clustermax-30-the-industry-standard","date":"2026-09-23","support":"ClusterMAX 3.0 summary: 'Nebius joins CoreWeave in the Platinum tier'; Google Cloud is placed in the Gold tier (full date shown)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://newsletter.semianalysis.com/p/clustermax-30-the-industry-standard","date":"2026-09-23","support":"SemiAnalysis describes 'capacity reservations backlogged for months' (Lambda section), so new GPU capacity is not available on demand (full date shown)."},{"url":"https://www.sec.gov/Archives/edgar/data/0001513845/000110465926094568/tm2622968d1_ex99-2.htm","date":"2026-08-12","support":"Nebius says 'Capacity remains a binding constraint on AI adoption' and it could sell all its 2027 capacity today (Q2 2026 letter, full date)."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/0001513845/000110465926094568/tm2622968d1_ex99-2.htm","date":"2026-08-12","support":"Nebius says 'Capacity remains a binding constraint on AI adoption' and it could sell all its 2027 capacity today (Q2 2026 letter, full date)."},{"url":"https://investors.coreweave.com/news/news-details/2026/CoreWeave-Reports-Strong-Second-Quarter-2026-Results/","date":"2026-08-11","support":"CoreWeave says active power reached 1.5 GW after adding nearly 500 MW in the quarter; contracted power about 3.7 GW (Q2 2026 release, full date)."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"nebius","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"gpu-capacity":{"alternatives":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/i/217033016/nebius","date":"2026-09-23","support":"ClusterMAX 3.0 rated 77 GPU clouds (executive summary): Platinum CoreWeave and Nebius; Gold Google Cloud and Oracle; Silver Azure, Lambda, Firmus, TensorWave, GMI; Bronze Crusoe. Many qualified providers sit at or near Nebius's quality."},{"url":"https://srgresearch.com/articles/neocloud-market-forecast-to-approach-400b-by-2031-driven-by-surging-ai-infrastructure-demand","date":"2026-04-02","support":"Synergy (2 Apr 2026): neocloud revenue exceeded $25B in 2025; leaders named are CoreWeave, Crusoe, Core Scientific, Lambda, Nebius, Nscale."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://assets.nebius.com/assets/4462517b-ce83-41f2-96ed-f2ac1bc06a05/SHLQ226%20(1).pdf-0","date":"2026-08-12","support":"Nebius says capacity remains a binding constraint and prepayment is the market standard for securing capacity in a supply-constrained market (letter, 12 Aug 2026)."},{"url":"https://www.microsoft.com/en-us/investor/events/fy-2026/earnings-fy-2026-q4","date":"2026-07-29","support":"Microsoft says it added about a gigawatt in one quarter, so peers add capacity at a pace that can cover a ~2.7% share within months to a year (our inference; call transcript, 29 Jul 2026)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/i/217033016/nebius","date":"2026-09-23","support":"ClusterMAX 3.0 puts CoreWeave on Platinum beside Nebius and Oracle and Google on Gold, so the best alternatives match or exceed Nebius's rated quality."},{"url":"https://storagenewsletter.com/2026/09/23/oracle-fiscal-1q27-financial-results","date":"2026-09-23","support":"Oracle reports RPO of $664B and over 300,000 GPUs delivered since the end of Q4 FY26, so larger alternatives exist (reported 23 Sep 2026)."}]}}}}},"lambda":{"a":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"lambda","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"gpu-capacity":{"alternatives":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/p/clustermax-30-the-industry-standard","date":"2026-09-23","support":"SemiAnalysis rates 77 GPU clouds; 'only 19 neoclouds globally achieve a Medallion rating', besides the hyperscalers (full date shown)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://newsletter.semianalysis.com/p/clustermax-30-the-industry-standard","date":"2026-09-23","support":"SemiAnalysis describes 'capacity reservations backlogged for months' (Lambda section), so new GPU capacity is not available on demand (full date shown)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://investors.coreweave.com/news/news-details/2026/CoreWeave-Reports-Strong-Second-Quarter-2026-Results/","date":"2026-08-11","support":"CoreWeave alone added nearly 500 MW in one quarter, more than Lambda's reported ~$1.5B a year of revenue implies (5LayerCake comparison; full date)."},{"url":"https://newsletter.semianalysis.com/p/clustermax-30-the-industry-standard","date":"2026-09-23","support":"ClusterMAX 3.0 summary: 'Nebius joins CoreWeave in the Platinum tier'; Google Cloud is placed in the Gold tier (full date shown)."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"lambda","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"gpu-capacity":{"alternatives":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/i/217033016/nebius","date":"2026-09-23","support":"ClusterMAX 3.0 rated 77 GPU clouds (executive summary): Platinum CoreWeave and Nebius; Gold Google Cloud and Oracle; Silver Azure, Lambda, Firmus, TensorWave, GMI; Bronze Crusoe. Lambda is Silver; many clouds rate at or above it."},{"url":"https://srgresearch.com/articles/neocloud-market-forecast-to-approach-400b-by-2031-driven-by-surging-ai-infrastructure-demand","date":"2026-04-02","support":"Synergy (2 Apr 2026) names Lambda among six neocloud leaders alongside CoreWeave, Crusoe, Core Scientific, Nebius and Nscale."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://lambda.ai/blog/lambda-announces-multibillion-dollar-agreement-with-microsoft-to-deploy-ai-infrastructure-powered-by-tens-of-thousands-of-nvidia-gpus","date":"2025-11-03","support":"Lambda says it signed a multibillion-dollar, multi-year Microsoft agreement for tens of thousands of NVIDIA GPUs, including GB300 NVL72 (3 Nov 2025)."},{"url":"https://assets.nebius.com/assets/4462517b-ce83-41f2-96ed-f2ac1bc06a05/SHLQ226%20(1).pdf-0","date":"2026-08-12","support":"Nebius says capacity is a binding constraint and prepayment now secures capacity in a supply-constrained market (letter, 12 Aug 2026)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/i/217033016/nebius","date":"2026-09-23","support":"Platinum CoreWeave and Nebius and Gold Oracle and Google are rated above Lambda's Silver, so alternatives are not weaker."},{"url":"https://newsletter.semianalysis.com/p/clustermax-20-the-industry-standard","date":"2025-11-06","support":"ClusterMAX 2.0: Lambda Silver; CoreWeave Platinum; Nebius, Oracle, Azure, Crusoe, Fluidstack Gold."}]}}}}},"baidu-ai-cloud":{"a":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"baidu-ai-cloud","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"china":{"alternatives":{"level":2,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"36Kr, citing Omdia: 2025 China AI cloud shares Alibaba 38.1%, Volcano 20.4%, Baidu 9.4%, Tencent 6.3%, eSurfing 5.9%; top five over 80% (full date)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://kr-asia.com/baidus-ai-revenue-share-holds-at-50-as-wall-street-funds-buy-in","date":"2026-08-19","support":"KrASIA: Baidu's GPU cloud revenue rose 283% and Kunlunxin has commercialised three chip generations (full date)."},{"url":"https://www.businessday.co.za/world/international-companies/2026-03-18-tencent-to-ramp-up-ai-investment-after-chip-curbs-slowed-spending/","date":"2026-03-18","support":"Reuters: Tencent said chip export restrictions held back its 2025 spending; capex fell short of internal targets on limited access to AI chips (full date)."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"36Kr: China cloud vendors' 'products and basic performance are gradually converging'; customers pick the cheapest when performance is close (full date)."},{"url":"https://kr-asia.com/baidus-ai-revenue-share-holds-at-50-as-wall-street-funds-buy-in","date":"2026-08-19","support":"KrASIA: Baidu's GPU cloud revenue rose 283% and Kunlunxin has commercialised three chip generations (full date)."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"baidu-ai-cloud","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"china":{"alternatives":{"level":1,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"Omdia via 36Kr (4 Aug 2026): five providers each hold over 5% of China's AI cloud and together over 80%; Huawei, China Mobile, Unicom and others also sell AI compute, so more than six alternatives exist."},{"url":"https://www.prnewswire.com/news-releases/baidu-announces-second-quarter-2026-results-302853860.html","date":"2026-08-18","support":"Baidu says AI Cloud Infra revenue was RMB 7.3B in Q2 2026 (+50% y/y) and GPU Cloud revenue rose 283% (company release, 18 Aug 2026)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.trendforce.com/news/?p=60480","date":"2026-07-09","support":"TrendForce (9 Jul 2026): Beijing may approve under 200,000 NVIDIA H200 chips, under half of what firms sought; Chinese AI developers are racing to build in-house chips to cope with shortages; inference is steered to domestic chips."},{"url":"https://www.prnewswire.com/news-releases/baidu-announces-second-quarter-2026-results-302853860.html","date":"2026-08-18","support":"Baidu's Q2 2026 capital expenditure was RMB 11.39B against RMB 3.8B a year earlier, showing peers can add capacity within quarters (company release cash-flow table, 18 Aug 2026)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"Alibaba (38.1%) and Volcano Engine (20.4%) are larger and fully capable AI clouds, so the best alternatives show no capability shortfall (Omdia via 36Kr, 4 Aug 2026)."}]}}}}},"iren":{"a":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"iren","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"gpu-capacity":{"alternatives":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/p/clustermax-30-the-industry-standard","date":"2026-09-23","support":"SemiAnalysis rates 77 GPU clouds; 'only 19 neoclouds globally achieve a Medallion rating', besides the hyperscalers (full date shown)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/0001878848/000187884826000051/irenreportsfy26results.htm","date":"2026-08-27","support":"IREN says its 2026 capacity is largely sold out and it has $1bn ARR operating today, with deliveries to Microsoft under way (FY26 release, full date)."},{"url":"https://newsletter.semianalysis.com/p/clustermax-30-the-industry-standard","date":"2026-09-23","support":"SemiAnalysis describes 'capacity reservations backlogged for months' (Lambda section), so new GPU capacity is not available on demand (full date shown)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://investors.coreweave.com/news/news-details/2026/CoreWeave-Reports-Strong-Second-Quarter-2026-Results/","date":"2026-08-11","support":"CoreWeave alone added nearly 500 MW of active power in one quarter, against IREN's $1bn operating ARR (5LayerCake comparison; full date)."},{"url":"https://www.sec.gov/Archives/edgar/data/0001878848/000187884826000051/irenreportsfy26results.htm","date":"2026-08-27","support":"IREN says its 2026 capacity is largely sold out and it has $1bn ARR operating today, with deliveries to Microsoft under way (FY26 release, full date)."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"iren","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"gpu-capacity":{"alternatives":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/i/217033016/nebius","date":"2026-09-23","support":"ClusterMAX 3.0 rated 77 GPU clouds (executive summary): Platinum CoreWeave and Nebius; Gold Google Cloud and Oracle; Silver Azure, Lambda, Firmus, TensorWave, GMI; Bronze Crusoe."},{"url":"https://srgresearch.com/articles/neocloud-market-forecast-to-approach-400b-by-2031-driven-by-surging-ai-infrastructure-demand","date":"2026-04-02","support":"Synergy (2 Apr 2026): neocloud revenue exceeded $25B in 2025, with six named leaders and demand outstripping hyperscale capacity."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1878848/000187884826000051/irenreportsfy26results.htm","date":"2026-08-27","support":"IREN says its 2026 capacity is largely sold out; Horizon 1 for Microsoft was delivered in Aug 2026, Horizons 2-4 targeted in Q4 2026 (FY26 release, 27 Aug 2026)."},{"url":"https://www.microsoft.com/en-us/investor/events/fy-2026/earnings-fy-2026-q4","date":"2026-07-29","support":"Microsoft says it added about a gigawatt in one quarter, so peers can cover IREN's ~1% share within months to a year (our inference; call transcript, 29 Jul 2026)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/i/217033016/nebius","date":"2026-09-23","support":"Platinum CoreWeave and Nebius and Gold Oracle and Google are rated above or alongside smaller GPU clouds; the best alternatives are far larger."},{"url":"https://investors.coreweave.com/news/news-details/2026/CoreWeave-Reports-Strong-Second-Quarter-2026-Results/","date":"2026-08-11","support":"CoreWeave says Q2 2026 revenue was $2,575M, backlog about $104B, active power 1.5 GW and contracted power about 3.7 GW (company release, 11 Aug 2026). That scale dwarfs IREN's roughly $1B of operating ARR (IREN release)."}]}}}}},"tencent-cloud":{"a":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"tencent-cloud","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"china":{"alternatives":{"level":2,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"36Kr, citing Omdia: 2025 China AI cloud shares Alibaba 38.1%, Volcano 20.4%, Baidu 9.4%, Tencent 6.3%, eSurfing 5.9%; top five over 80% (full date)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.businessday.co.za/world/international-companies/2026-03-18-tencent-to-ramp-up-ai-investment-after-chip-curbs-slowed-spending/","date":"2026-03-18","support":"Reuters: Tencent said chip export restrictions held back its 2025 spending; capex fell short of internal targets on limited access to AI chips (full date)."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"36Kr: China cloud vendors' 'products and basic performance are gradually converging'; customers pick the cheapest when performance is close (full date)."},{"url":"https://technode.com/2025/11/26/alibaba-ceo-says-ai-bubble-unlikely-in-next-three-years/","date":"2025-11-26","support":"Alibaba CEO Eddie Wu says AI resources will remain in short supply, with no bubble likely in three years (TechNode, full date)."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"tencent-cloud","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"china":{"alternatives":{"level":1,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"Omdia via 36Kr (4 Aug 2026): five providers each hold over 5% of China's AI cloud and together over 80%; Huawei, China Mobile, Unicom and others also sell AI compute, so more than six alternatives exist."},{"url":"https://eu.36kr.com/en/p/3936648709815425","date":"2026-08-13","support":"Tencent says new capacity goes first to its own models and applications, with external leasing weighed alongside, so its external supply is discretionary (36Kr, 13 Aug 2026)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.trendforce.com/news/?p=60480","date":"2026-07-09","support":"TrendForce (9 Jul 2026): Beijing may approve under 200,000 NVIDIA H200 chips, under half of what firms sought; Chinese AI developers are racing to build in-house chips to cope with shortages; inference is steered to domestic chips."},{"url":"https://eu.36kr.com/en/p/3936648709815425","date":"2026-08-13","support":"Tencent expects enough GPU capacity by the end of the year and next to support bare-metal GPU leasing and MaaS (36Kr, 13 Aug 2026)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"Alibaba (38.1%) and Volcano Engine (20.4%) are larger and fully capable AI clouds, so the best alternatives show no capability shortfall (Omdia via 36Kr, 4 Aug 2026)."}]}}}}},"huawei-cloud":{"a":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"huawei-cloud","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"china":{"alternatives":{"level":2,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"36Kr, citing Omdia: 2025 China AI cloud shares Alibaba 38.1%, Volcano 20.4%, Baidu 9.4%, Tencent 6.3%, eSurfing 5.9%; top five over 80% (full date)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.businessday.co.za/world/international-companies/2026-03-18-tencent-to-ramp-up-ai-investment-after-chip-curbs-slowed-spending/","date":"2026-03-18","support":"Reuters: Tencent said chip export restrictions held back its 2025 spending; capex fell short of internal targets on limited access to AI chips (full date)."},{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"36Kr: China cloud vendors' 'products and basic performance are gradually converging'; customers pick the cheapest when performance is close (full date)."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.huaweicloud.com/eu/news/20250919133255709.html","date":"2025-09-19","support":"The company says AI Compute Service customers rose from 321 in 2024 to 1,805, on CloudMatrix384 supernodes (press release dated 2025-09-19; page shows 2026-03-20)."},{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"36Kr: China cloud vendors' 'products and basic performance are gradually converging'; customers pick the cheapest when performance is close (full date)."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"huawei-cloud","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"china":{"alternatives":{"level":1,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"36Kr, citing Omdia data released 19 May: China's AI cloud market was 56.7B yuan in 2025; Alibaba 38.1%, Volcano Engine 20.4%, Baidu 9.4%, Tencent 6.3%, eSurfing (China Telecom) 5.9%; top five over 80%. Five larger or similar AI clouds plus state-owned carriers serve the same customers."},{"url":"https://www.prnewswire.com/news-releases/baidu-announces-second-quarter-2026-results-302853860.html","date":"2026-08-18","support":"Baidu says AI Cloud Infra revenue was RMB 7.3B and GPU Cloud revenue rose 283% in Q2 2026, another domestic cloud serving AI compute demand (company release, 18 Aug 2026)."}]},"switchingTime":{"level":2,"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 via Gulf News: Huawei plans about 600,000 910C Ascend chips in 2026, roughly double 2025, limited by yields and HBM (page last updated 30 Sep 2025)."},{"url":"https://www.trendforce.com/news/?p=60480","date":"2026-07-09","support":"TrendForce (9 Jul 2026): Beijing may approve under 200,000 NVIDIA H200 chips, under half of what firms sought; Chinese AI developers are racing to build in-house chips to cope with shortages; inference is steered to domestic chips."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.huaweicloud.com/eu/news/20250919133255709.html","date":"2025-09-19","support":"Huawei says CloudMatrix384 inference per card is 3-4x NVIDIA's H20 and supernodes will scale from 384 to 8,192 cards (company claim, 19 Sep 2025)."},{"url":"https://www.trendforce.com/news/?p=60480","date":"2026-07-09","support":"TrendForce: Beijing steers inference toward domestic processors including Huawei's, so rivals running imported chips cover less of this demand (9 Jul 2026)."}]}}}}},"china-telecom-tianyi-cloud":{"a":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"china-telecom-tianyi-cloud","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"china":{"alternatives":{"level":2,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"36Kr, citing Omdia: 2025 China AI cloud shares Alibaba 38.1%, Volcano 20.4%, Baidu 9.4%, Tencent 6.3%, eSurfing 5.9%; top five over 80% (full date)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.businessday.co.za/world/international-companies/2026-03-18-tencent-to-ramp-up-ai-investment-after-chip-curbs-slowed-spending/","date":"2026-03-18","support":"Reuters: Tencent said chip export restrictions held back its 2025 spending; capex fell short of internal targets on limited access to AI chips (full date)."},{"url":"https://www.caixinglobal.com/2026-03-25/china-telecom-to-boost-ai-spending-amid-capex-cut-and-slowing-growth-102426994.html","date":"2026-03-25","support":"Caixin: China Telecom's computing capacity reached 46 EFLOPS in 2025 and it is sharply raising AI computing spending for 2026 (full date)."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.caixinglobal.com/2026-03-20/china-unicom-slashes-spending-pivots-harder-to-ai-102424845.html","date":"2026-03-20","support":"Caixin: China Unicom's intelligent computing capacity reached 45 EFLOPS at end-2025 and over a third of 2026 capex goes to computing (full date)."},{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"36Kr: China cloud vendors' 'products and basic performance are gradually converging'; customers pick the cheapest when performance is close (full date)."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"china-telecom-tianyi-cloud","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"china":{"alternatives":{"level":1,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"Omdia via 36Kr (4 Aug 2026): five providers each hold over 5% of China's AI cloud and together over 80%; Huawei, China Mobile, Unicom and others also sell AI compute, so more than six alternatives exist."},{"url":"https://www.caixinglobal.com/2026-03-25/china-telecom-to-boost-ai-spending-amid-capex-cut-and-slowing-growth-102426994.html","date":"2026-03-25","support":"Caixin (25 Mar 2026): China Telecom had 46 EFLOPS of computing by 2025, AI revenue of 12.3B yuan, and 2026 computing capex up 26%."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.trendforce.com/news/?p=60480","date":"2026-07-09","support":"TrendForce (9 Jul 2026): Beijing may approve under 200,000 NVIDIA H200 chips, under half of what firms sought; Chinese AI developers are racing to build in-house chips to cope with shortages; inference is steered to domestic chips."},{"url":"https://www.caixinglobal.com/2026-03-25/china-telecom-to-boost-ai-spending-amid-capex-cut-and-slowing-growth-102426994.html","date":"2026-03-25","support":"China Telecom's 2026 capex is 73B yuan (-9.2%) with 35% for computing, so peers of this size add capacity over a year (our inference; Caixin, 25 Mar 2026)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"Alibaba (38.1%) and Volcano Engine (20.4%) are larger and fully capable AI clouds, so the best alternatives show no capability shortfall (Omdia via 36Kr, 4 Aug 2026)."}]}}}}},"crusoe":{},"together-ai":{},"kingsoft-cloud":{"a":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"kingsoft-cloud","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"china":{"alternatives":{"level":1,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"Omdia via 36Kr: five providers each hold 5.9% or more, far above Kingsoft Cloud's estimated 1.5% (full date)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.businessday.co.za/world/international-companies/2026-03-18-tencent-to-ramp-up-ai-investment-after-chip-curbs-slowed-spending/","date":"2026-03-18","support":"Reuters: Tencent said chip export restrictions held back its 2025 spending; capex fell short of internal targets on limited access to AI chips (full date)."},{"url":"https://www.sec.gov/Archives/edgar/data/0001795589/000110465926098490/tm2623439d2_ex99-1.htm","date":"undated","support":"The company says AI cloud gross billings grew 82% and Q2 capex was RMB3.3B; 5LayerCake reads this as a small fleet beside the leaders (release shows no date; call set for 19 Aug 2026)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"36Kr: China cloud vendors' 'products and basic performance are gradually converging'; customers pick the cheapest when performance is close (full date)."},{"url":"https://www.sec.gov/Archives/edgar/data/0001795589/000110465926098490/tm2623439d2_ex99-1.htm","date":"undated","support":"The company says AI cloud gross billings grew 82% and Q2 capex was RMB3.3B; 5LayerCake reads this as a small fleet beside the leaders (release shows no date; call set for 19 Aug 2026)."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"kingsoft-cloud","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"china":{"alternatives":{"level":1,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"Omdia via 36Kr (4 Aug 2026): five providers each hold over 5% of China's AI cloud and together over 80%; Huawei, China Mobile, Unicom and others also sell AI compute, so more than six alternatives exist."},{"url":"https://www.sec.gov/Archives/edgar/data/0001795589/000110465926098490/tm2623439d2_ex99-1.htm","date":"2026-08","support":"Kingsoft Cloud's Q2 2026 release: AI cloud gross billings RMB 1,327M (+82% y/y), 56% of public cloud revenue; release date not shown (month precision, call on 19 Aug 2026)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.trendforce.com/news/?p=60480","date":"2026-07-09","support":"TrendForce (9 Jul 2026): Beijing may approve under 200,000 NVIDIA H200 chips, under half of what firms sought; Chinese AI developers are racing to build in-house chips to cope with shortages; inference is steered to domestic chips."},{"url":"https://www.sec.gov/Archives/edgar/data/0001795589/000110465926098490/tm2623439d2_ex99-1.htm","date":"2026-08","support":"Kingsoft Cloud's Q2 2026 capex including leased assets was RMB 3.3B; a 1.5% market share is a small volume for peers to absorb (our inference; release, month precision)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"Alibaba (38.1%) and Volcano Engine (20.4%) are larger and fully capable AI clouds, so the best alternatives show no capability shortfall (Omdia via 36Kr, 4 Aug 2026)."}]}}}}},"core42":{},"china-mobile-cloud":{"a":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"china-mobile-cloud","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"china":{"alternatives":{"level":1,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"Omdia via 36Kr: five providers each hold more than China Mobile's estimated 3%, and Huawei Cloud is also larger (full date)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.businessday.co.za/world/international-companies/2026-03-18-tencent-to-ramp-up-ai-investment-after-chip-curbs-slowed-spending/","date":"2026-03-18","support":"Reuters: Tencent said chip export restrictions held back its 2025 spending; capex fell short of internal targets on limited access to AI chips (full date)."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.caixinglobal.com/2026-03-25/china-telecom-to-boost-ai-spending-amid-capex-cut-and-slowing-growth-102426994.html","date":"2026-03-25","support":"Caixin: China Telecom's computing capacity reached 46 EFLOPS in 2025 and it is sharply raising AI computing spending for 2026 (full date)."},{"url":"https://technode.com/2025/11/26/alibaba-ceo-says-ai-bubble-unlikely-in-next-three-years/","date":"2025-11-26","support":"Alibaba CEO Eddie Wu says AI resources will remain in short supply, with no bubble likely in three years (TechNode, full date)."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"china-mobile-cloud","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"china":{"alternatives":{"level":1,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"Omdia via 36Kr (4 Aug 2026): five providers each hold over 5% of China's AI cloud and together over 80%; Huawei, China Mobile, Unicom and others also sell AI compute, so more than six alternatives exist."},{"url":"https://www1.hkexnews.hk/listedco/listconews/sehk/2026/0326/2026032600468.pdf","date":"2026-03-26","support":"China Mobile's 2025 results announcement: intelligent computing capacity 92.5 EFLOPS (FP16) and intelligent computing services revenue up 279% (26 Mar 2026)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.trendforce.com/news/?p=60480","date":"2026-07-09","support":"TrendForce (9 Jul 2026): Beijing may approve under 200,000 NVIDIA H200 chips, under half of what firms sought; Chinese AI developers are racing to build in-house chips to cope with shortages; inference is steered to domestic chips."},{"url":"https://www1.hkexnews.hk/listedco/listconews/sehk/2026/0326/2026032600468.pdf","date":"2026-03-26","support":"China Mobile's computing services revenue was RMB 89.8B in 2025; capacity at this scale is built over quarters (our inference; results announcement, 26 Mar 2026)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"Alibaba (38.1%) and Volcano Engine (20.4%) are larger and fully capable AI clouds, so the best alternatives show no capability shortfall (Omdia via 36Kr, 4 Aug 2026)."}]}}}}},"fluidstack":{},"modal":{},"voltage-park":{},"nscale":{},"china-unicom-cloud":{"a":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"china-unicom-cloud","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"china":{"alternatives":{"level":1,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"Omdia via 36Kr: five providers each hold 5.9% or more, far above China Unicom's estimated 1.8% (full date)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.businessday.co.za/world/international-companies/2026-03-18-tencent-to-ramp-up-ai-investment-after-chip-curbs-slowed-spending/","date":"2026-03-18","support":"Reuters: Tencent said chip export restrictions held back its 2025 spending; capex fell short of internal targets on limited access to AI chips (full date)."},{"url":"https://www.caixinglobal.com/2026-03-20/china-unicom-slashes-spending-pivots-harder-to-ai-102424845.html","date":"2026-03-20","support":"Caixin: China Unicom's intelligent computing capacity reached 45 EFLOPS at end-2025 and over a third of 2026 capex goes to computing (full date)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.caixinglobal.com/2026-03-25/china-telecom-to-boost-ai-spending-amid-capex-cut-and-slowing-growth-102426994.html","date":"2026-03-25","support":"Caixin: China Telecom's computing capacity reached 46 EFLOPS in 2025 and it is sharply raising AI computing spending for 2026 (full date)."},{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"36Kr: China cloud vendors' 'products and basic performance are gradually converging'; customers pick the cheapest when performance is close (full date)."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"china-unicom-cloud","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"china":{"alternatives":{"level":1,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"Omdia via 36Kr (4 Aug 2026): five providers each hold over 5% of China's AI cloud and together over 80%; Huawei, China Mobile, Unicom and others also sell AI compute, so more than six alternatives exist."},{"url":"https://www.rcrwireless.com/20260324/5g/china-unicom-ai","date":"2026-03-24","support":"RCR Wireless (24 Mar 2026): China Unicom's computing capacity reached 45 EFLOPS, AI revenue rose 147%, and over 35% of 2026 capex is for computing."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.trendforce.com/news/?p=60480","date":"2026-07-09","support":"TrendForce (9 Jul 2026): Beijing may approve under 200,000 NVIDIA H200 chips, under half of what firms sought; Chinese AI developers are racing to build in-house chips to cope with shortages; inference is steered to domestic chips."},{"url":"https://www.rcrwireless.com/20260324/5g/china-unicom-ai","date":"2026-03-24","support":"China Unicom's 2026 capex is about CNY 50B with over 35% on computing, so a 1.8% share is within one year's peer build-out (our inference; RCR Wireless, 24 Mar 2026)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"Alibaba (38.1%) and Volcano Engine (20.4%) are larger and fully capable AI clouds, so the best alternatives show no capability shortfall (Omdia via 36Kr, 4 Aug 2026)."}]}}}}},"runpod":{},"sensetime":{"a":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"sensetime","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"china":{"alternatives":{"level":1,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"Omdia via 36Kr: five providers each hold 5.9% or more, far above SenseTime's estimated 1.5% (full date)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://sensetime.com/en/news/2026H1-results","date":"2026-08-26","support":"The company says SenseCore has 48,000 PFLOPS deployed and its Token Factory serves four external foundation model providers (full date)."},{"url":"https://www.businessday.co.za/world/international-companies/2026-03-18-tencent-to-ramp-up-ai-investment-after-chip-curbs-slowed-spending/","date":"2026-03-18","support":"Reuters: Tencent said chip export restrictions held back its 2025 spending; capex fell short of internal targets on limited access to AI chips (full date)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.caixinglobal.com/2026-03-25/china-telecom-to-boost-ai-spending-amid-capex-cut-and-slowing-growth-102426994.html","date":"2026-03-25","support":"Caixin: China Telecom's computing capacity reached 46 EFLOPS in 2025 and it is sharply raising AI computing spending for 2026 (full date)."},{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"36Kr: China cloud vendors' 'products and basic performance are gradually converging'; customers pick the cheapest when performance is close (full date)."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"infrastructure-cloud","entityId":"sensetime","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"china":{"alternatives":{"level":1,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"Omdia via 36Kr (4 Aug 2026): five providers each hold over 5% of China's AI cloud and together over 80%; Huawei, China Mobile, Unicom and others also sell AI compute, so more than six alternatives exist."},{"url":"https://sensetime.com/en/news/2026H1-results","date":"2026-08-26","support":"SenseTime says SenseCore had 48,000 petaflops deployed and served four external foundation-model providers (H1 2026 results, 26 Aug 2026)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.trendforce.com/news/?p=60480","date":"2026-07-09","support":"TrendForce (9 Jul 2026): Beijing may approve under 200,000 NVIDIA H200 chips, under half of what firms sought; Chinese AI developers are racing to build in-house chips to cope with shortages; inference is steered to domestic chips."},{"url":"https://sensetime.com/en/news/2026H1-results","date":"2026-08-26","support":"SenseTime says SenseCore served 2.4 trillion tokens a day in July 2026; larger clouds can absorb this workload (our inference; H1 2026 results, 26 Aug 2026)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://eu.36kr.com/en/p/3924679520647302","date":"2026-08-04","support":"Alibaba (38.1%) and Volcano Engine (20.4%) are larger and fully capable AI clouds, so the best alternatives show no capability shortfall (Omdia via 36Kr, 4 Aug 2026)."}]}}}}},"nvidia":{},"rumble":{},"sakura-internet":{},"mistral":{},"naver-cloud":{},"tensorwave":{},"vultr":{},"gmo-internet":{},"yotta":{},"nexgen-cloud":{},"softbank-corp":{},"kddi":{},"ntt":{}},"validation":[]}