{"schema":"5lc-index-ledger/1","key":"models-inference-serving","layer":"models","sublayer":"inference-serving","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":"microsoft-intelligent-cloud","slot":"importance","since":"2026-10-22"},{"entityId":"aws","slot":"importance","since":"2026-10-22"},{"entityId":"google-cloud","slot":"importance","since":"2026-10-22"},{"entityId":"fireworks-ai","slot":"importance","since":"2026-10-22"},{"entityId":"fal","slot":"importance","since":"2026-10-22"},{"entityId":"baseten","slot":"importance","since":"2026-10-22"},{"entityId":"cerebras-inference","slot":"importance","since":"2026-10-22"},{"entityId":"together-ai","slot":"importance","since":"2026-10-22"},{"entityId":"modal","slot":"importance","since":"2026-10-22"},{"entityId":"volcano-engine","slot":"importance","since":"2026-10-22"}],"pending":{},"lastRun":"2026-10-22"},"card":{"schema":"5lc-index-card/1","key":"models-inference-serving","layer":"models","sublayer":"inference-serving","version":1,"scope":"Running models for others, sold per token, per request or as dedicated deployments (including deployments billed by the hour when the provider runs and manages the model), when the provider did not develop the model. Any model served by a cloud or data platform (Amazon, Microsoft, Google, Oracle, Databricks, Snowflake and others) counts in Cloud model platforms; independent providers count in the language-model or the image, video and audio part, and all serving in mainland China in its own part. A developer selling its own models counts in Frontier labs, Open-weight or Specialist models; bare accelerator instances and capacity a model developer buys to serve its own users count in Infrastructure · Cloud; routing gateways that do not run models are outside this sublayer.","outputMeasure":"Revenue providers keep for running others' models, after any share paid to the model's developer.","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","hardToReplace":"Most serving runs on standard chips and open software, so customers can move within weeks. Harder to replace are clouds holding enterprise contracts, data and security approvals, and the few providers with their own fast inference chips.","functions":[{"id":"cloud-model-platforms","name":"Cloud model platforms","why":"Amazon, Microsoft, Google, Oracle, Databricks, Snowflake and other platforms serve other developers' models inside them.","weight":{"low":0.45,"central":0.687,"high":0.852},"weightSource":{"url":"https://www.khaleejtimes.com/business/anthropic-ipo-prospectus-lays-bare-deep-dependence-on-big-tech-partners","title":"Anthropic IPO prospectus lays bare deep dependence on Big Tech partners","publisher":"Reuters (published by Khaleej Times)","publishedDate":"2026-09-30","accessedDate":"2026-10-22","tier":"T3","locator":"Anthropic's confidential IPO filing: 2025 revenue nearly $4.6bn; sales through cloud marketplaces about $2.16bn (47%); distribution fees about $351m, roughly 16 cents per dollar sold through the platforms"},"shareData":"Platform disclosures, developers' channel splits and fees, and press estimates; never a platform's total AI run-rate"},{"id":"independent-llm-serving","name":"Independent inference providers","why":"Specialists serving open and custom language models, often faster or cheaper.","weight":{"low":0.103,"central":0.226,"high":0.423},"weightSource":{"url":"https://fireworks.ai/blog/series-d-announcement","title":"Fireworks Secures $1.5 Billion in Series D Funding","publisher":"Fireworks AI","publishedDate":"2026-07-15","accessedDate":"2026-10-22","tier":"T2","locator":"Annualized revenue run rate above $1 billion; more than 40 trillion tokens served daily, over 95% from customer-specialised models"},"shareData":"Company run-rate statements and filings, less GPU rental and lab-bought capacity"},{"id":"media-model-serving","name":"Image, video and audio model hosts","why":"Independent hosts that run image, video and sound models for app builders.","weight":{"low":0.009,"central":0.028,"high":0.092},"weightSource":{"url":"https://techcrunch.com/2025/12/09/fal-nabs-140m-in-fresh-funding-led-by-sequoia-tripling-valuation-to-4-5b/","title":"Fal nabs $140M in fresh funding led by Sequoia, tripling valuation to $4.5B","publisher":"TechCrunch","publishedDate":"2025-12-09","accessedDate":"2026-10-22","tier":"T3","locator":"Citing Bloomberg: fal had surpassed $200 million in revenue as of October 2025"},"shareData":"Company revenue statements net of payments to model developers"},{"id":"china-serving","name":"Model serving in mainland China","why":"Chinese clouds and specialists serving others' models in a separate market.","weight":{"low":0.02,"central":0.059,"high":0.159},"weightSource":{"url":"https://www.idc.com/?p=114742","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","accessedDate":"2026-10-22","tier":"T4","locator":"China public-cloud MaaS revenue RMB 3.07bn in 2025, 2026 forecast about RMB 18.6bn; call-volume ranking Volcano Engine, Alibaba Cloud, Baidu AI Cloud, SiliconFlow, China Mobile Cloud"},"shareData":"IDC platform shares, third-party models only"}],"risingFloorPct":1,"validationCaseIds":[],"knownGaps":["Amazon, Microsoft and Google do not disclose model-serving revenue; the cloud weight rests on Anthropic's 2025 channel split and 16% platform fee, carried to 2026, plus a 5LayerCake estimate for OpenAI models sold through Microsoft Azure (US$3–12bn a year gross, built from reported revenue-share payments).","Microsoft's reported 20% revenue share with OpenAI is not confirmed by either company.","Sales through cloud platforms are counted in full for the model's developer in Frontier labs, and the platform's retained part is counted again here.","Most independent providers are private and report run-rates that mix serving with GPU rental, training, batch jobs or sandboxes; Baseten publishes growth but no revenue figure, so its size is an estimate.","fal and similar hosts report gross revenue, including payments to the developers of the proprietary models they host; their net figures are estimates.","China's MaaS totals mix platforms' own models with third-party models; the third-party portion is estimated (20 to 45%)."],"approval":{"approver":"Claude (delegated by Dachi)","date":"2026-10-22","sha256":"bc5a0bef51366b6e63164958f7ba76631ed8995b9c353fe8b44b8fcc56de9be8","delegatedOn":"2026-10-22"}},"result":{"schema":"5lc-index-result/1","key":"models-inference-serving","asOf":"2026-10-22","methodVersion":"1.1","methodSha256":"3cabed6b38514c263d8b86a494e8cc19cdbd0b2a1013543dc024192c49f7a6a6","cardSha256":"bc5a0bef51366b6e63164958f7ba76631ed8995b9c353fe8b44b8fcc56de9be8","validationSha256":"409931bb458f38b02342c9bf43e61eab299e47d8679912af2c96074ec556f17e","inputsSha256":"dd2b19babc02e03919e82f48732dddab7f385344818fcd369901246b80d102b3","unscreened":[],"companies":[{"entityId":"microsoft-intelligent-cloud","name":"Microsoft Azure (Foundry)","reachPct":35.251344,"importance":19.784147896,"importanceRank":1,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":51.312,"shareLowPct":33.575,"shareHighPct":68.538,"stale":false,"offBasis":false,"level":2.16666666667,"years":0.5612310241559831,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china-serving","weight":0.059,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{},"mc":{"medianRank":1,"rankP10":1,"rankP90":1,"pTop8":1,"pSelected":1,"stability":"firm"},"baselineRevenueRank":2},{"entityId":"aws","name":"Amazon Web Services (AWS)","reachPct":20.076888,"importance":11.2677724141,"importanceRank":2,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":29.224,"shareLowPct":19.066,"shareHighPct":41.372,"stale":false,"offBasis":false,"level":2.16666666667,"years":0.5612310241559831,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china-serving","weight":0.059,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{},"mc":{"medianRank":2,"rankP10":2,"rankP90":2,"pTop8":1,"pSelected":1,"stability":"firm"},"baselineRevenueRank":1},{"entityId":"google-cloud","name":"Google Cloud (Vertex AI)","reachPct":9.622809,"importance":4.28647410149,"importanceRank":3,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":14.007,"shareLowPct":7.127,"shareHighPct":25.716,"stale":false,"offBasis":false,"level":1.83333333333,"years":0.4454493590691404,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china-serving","weight":0.059,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{},"mc":{"medianRank":3,"rankP10":3,"rankP90":4,"pTop8":1,"pSelected":1,"stability":"firm"},"baselineRevenueRank":3},{"entityId":"fireworks-ai","name":"Fireworks AI","reachPct":6.360772,"importance":2.24758513164,"importanceRank":4,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":28.09,"shareLowPct":27.066,"shareHighPct":35.354,"stale":false,"offBasis":false,"level":1.5,"years":0.3535533905932738,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0.444,"shareLowPct":0,"shareHighPct":6.276,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"china-serving","weight":0.059,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{},"mc":{"medianRank":4,"rankP10":3,"rankP90":5,"pTop8":1,"pSelected":1,"stability":"firm"},"baselineRevenueRank":11},{"entityId":"fal","name":"fal","reachPct":2.066316,"importance":1.45965976982,"importanceRank":5,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":0.014,"shareLowPct":0,"shareHighPct":0.28,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"media-model-serving","weight":0.028,"sharePct":73.684,"shareLowPct":70.588,"shareHighPct":87.805,"stale":false,"offBasis":false,"level":2.5,"years":0.7071067811865475,"disagreement":false},{"id":"china-serving","weight":0.059,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{"growth":66.6666666667},"mc":{"medianRank":5,"rankP10":4,"rankP90":7,"pTop8":0.9362,"pSelected":0.9862,"stability":"firm"},"baselineRevenueRank":16},{"entityId":"baseten","name":"Baseten","reachPct":3.630918,"importance":1.14285967003,"importanceRank":6,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":16.011,"shareLowPct":8.282,"shareHighPct":25.063,"stale":false,"offBasis":false,"level":1.33333333333,"years":0.3149802624729905,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0.444,"shareLowPct":0,"shareHighPct":6.276,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"china-serving","weight":0.059,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{"growth":66.6666666667},"mc":{"medianRank":6,"rankP10":5,"rankP90":9,"pTop8":0.8806,"pSelected":0.9772,"stability":"firm"},"baselineRevenueRank":14},{"entityId":"cerebras-inference","name":"Cerebras Inference","reachPct":1.59008,"importance":1.00050671419,"importanceRank":7,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":7.022,"shareLowPct":4.335,"shareHighPct":9.563,"stale":false,"offBasis":false,"level":2.33333333333,"years":0.6299605249459811,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0.111,"shareLowPct":0,"shareHighPct":2.176,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"china-serving","weight":0.059,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{"growth":0},"mc":{"medianRank":8,"rankP10":6,"rankP90":10,"pTop8":0.6684,"pSelected":0.926,"stability":"firm"},"baselineRevenueRank":15},{"entityId":"together-ai","name":"Together AI","reachPct":2.881532,"importance":0.807840496422,"importanceRank":8,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":12.64,"shareLowPct":7.164,"shareHighPct":18.799,"stale":false,"offBasis":false,"level":1.16666666667,"years":0.2806155120779916,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0.889,"shareLowPct":0,"shareHighPct":10.081,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"china-serving","weight":0.059,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{},"mc":{"medianRank":8,"rankP10":6,"rankP90":10,"pTop8":0.5971,"pSelected":0.9054,"stability":"firm"},"baselineRevenueRank":13},{"entityId":"modal","name":"Modal","reachPct":2.500722,"importance":0.786060583239,"importanceRank":9,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":10.955,"shareLowPct":8.116,"shareHighPct":13.388,"stale":false,"offBasis":false,"level":1.33333333333,"years":0.3149802624729905,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0.889,"shareLowPct":0,"shareHighPct":11.858,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"china-serving","weight":0.059,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{"growth":66.6666666667},"mc":{"medianRank":9,"rankP10":7,"rankP90":11,"pTop8":0.3617,"pSelected":0.8603,"stability":"likely"},"baselineRevenueRank":12},{"entityId":"volcano-engine","name":"Volcano Engine (incl. BytePlus)","reachPct":1.302204,"importance":0.507498475848,"importanceRank":10,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0.092,"shareLowPct":0,"shareHighPct":0.46,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"independent-llm-serving","weight":0.226,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china-serving","weight":0.059,"sharePct":21,"shareLowPct":9,"shareHighPct":45,"stale":false,"offBasis":false,"level":1.66666666667,"years":0.3968502629929668,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{},"mc":{"medianRank":10,"rankP10":6,"rankP90":13,"pTop8":0.3176,"pSelected":0.5616,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"oracle-cloud-infrastructure","name":"Oracle Cloud Infrastructure (OCI)","reachPct":1.269576,"importance":0.399891381709,"importanceRank":11,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":1.848,"shareLowPct":0.375,"shareHighPct":5.348,"stale":false,"offBasis":false,"level":1.33333333333,"years":0.3149802624729905,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china-serving","weight":0.059,"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":9,"rankP90":17,"pTop8":0.0876,"pSelected":0.2378,"stability":"borderline"},"baselineRevenueRank":4},{"entityId":"groq","name":"Groq","reachPct":0.958354,"importance":0.338185618213,"importanceRank":12,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":4.213,"shareLowPct":1.445,"shareHighPct":12.788,"stale":false,"offBasis":false,"level":1.5,"years":0.3535533905932738,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0.222,"shareLowPct":0,"shareHighPct":4.264,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"china-serving","weight":0.059,"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":9,"rankP90":16,"pTop8":0.0674,"pSelected":0.271,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"databricks","name":"Databricks","reachPct":0.952182,"importance":0.336647174562,"importanceRank":13,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":1.386,"shareLowPct":0.373,"shareHighPct":3.613,"stale":false,"offBasis":false,"level":1.5,"years":0.3535533905932738,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china-serving","weight":0.059,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":13,"rankP10":10,"rankP90":18,"pTop8":0.0471,"pSelected":0.151,"stability":"borderline"},"baselineRevenueRank":6},{"entityId":"snowflake","name":"Snowflake","reachPct":0.761883,"importance":0.269366317885,"importanceRank":14,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":1.109,"shareLowPct":0.28,"shareHighPct":2.727,"stale":false,"offBasis":false,"level":1.5,"years":0.3535533905932738,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china-serving","weight":0.059,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":15,"rankP10":12,"rankP90":23,"pTop8":0.0051,"pSelected":0.0255,"stability":"borderline"},"baselineRevenueRank":7},{"entityId":"alibaba-cloud","name":"Alibaba Cloud (Model Studio)","reachPct":0.835095,"importance":0.230449532551,"importanceRank":15,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0.185,"shareLowPct":0,"shareHighPct":0.735,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"independent-llm-serving","weight":0.226,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china-serving","weight":0.059,"sharePct":12,"shareLowPct":5,"shareHighPct":29,"stale":false,"offBasis":false,"level":1.16666666667,"years":0.2806155120779916,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":13,"rankP10":11,"rankP90":17,"pTop8":0.031,"pSelected":0.0918,"stability":"borderline"},"baselineRevenueRank":5},{"entityId":"deepinfra","name":"DeepInfra","reachPct":0.58376,"importance":0.14594,"importanceRank":16,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":2.528,"shareLowPct":1.42,"shareHighPct":2.937,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0.444,"shareLowPct":0,"shareHighPct":6.276,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"china-serving","weight":0.059,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":21,"rankP10":17,"rankP90":26,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":19},{"entityId":"digitalocean","name":"DigitalOcean","reachPct":0.51093,"importance":0.1277325,"importanceRank":17,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":2.247,"shareLowPct":0.855,"shareHighPct":4.396,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0.111,"shareLowPct":0,"shareHighPct":2.176,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"china-serving","weight":0.059,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":21,"rankP10":16,"rankP90":29,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":10},{"entityId":"baidu-ai-cloud","name":"Baidu AI Cloud (Qianfan)","reachPct":0.472,"importance":0.118,"importanceRank":18,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china-serving","weight":0.059,"sharePct":8,"shareLowPct":2,"shareHighPct":19,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":20,"rankP10":15,"rankP90":30,"pTop8":0,"pSelected":0.0015,"stability":"borderline"},"baselineRevenueRank":8},{"entityId":"china-mobile-cloud","name":"China Mobile Cloud","reachPct":0.472,"importance":0.118,"importanceRank":19,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china-serving","weight":0.059,"sharePct":8,"shareLowPct":3,"shareHighPct":21,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":19,"rankP10":14,"rankP90":29,"pTop8":0.0002,"pSelected":0.0028,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"huawei-cloud","name":"Huawei Cloud","reachPct":0.413,"importance":0.10325,"importanceRank":20,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china-serving","weight":0.059,"sharePct":7,"shareLowPct":2,"shareHighPct":16,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":22,"rankP10":16,"rankP90":32,"pTop8":0,"pSelected":0.0003,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"tencent-cloud","name":"Tencent Cloud","reachPct":0.413,"importance":0.10325,"importanceRank":21,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china-serving","weight":0.059,"sharePct":7,"shareLowPct":2,"shareHighPct":16,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":22,"rankP10":16,"rankP90":32,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"coreweave","name":"CoreWeave (W&B Inference)","reachPct":0.383918,"importance":0.0959795,"importanceRank":22,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":1.685,"shareLowPct":1.13,"shareHighPct":2.778,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0.111,"shareLowPct":0,"shareHighPct":2.176,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"china-serving","weight":0.059,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":26,"rankP10":20,"rankP90":31,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"crusoe","name":"Crusoe (Managed Inference)","reachPct":0.383918,"importance":0.0959795,"importanceRank":23,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":1.685,"shareLowPct":0.85,"shareHighPct":2.778,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0.111,"shareLowPct":0,"shareHighPct":2.176,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"china-serving","weight":0.059,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":26,"rankP10":21,"rankP90":32,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"cloudflare-workers-ai","name":"Workers AI","reachPct":0.378456,"importance":0.094614,"importanceRank":24,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":1.124,"shareLowPct":0.283,"shareHighPct":3.297,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":4.444,"shareLowPct":2.273,"shareHighPct":15.686,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"china-serving","weight":0.059,"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":15,"rankP90":28,"pTop8":0,"pSelected":0.0001,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"hugging-face","name":"Hugging Face","reachPct":0.329736,"importance":0.082434,"importanceRank":25,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":1.404,"shareLowPct":0.284,"shareHighPct":3.306,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0.444,"shareLowPct":0,"shareHighPct":6.276,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"china-serving","weight":0.059,"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":18,"rankP90":33,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":17},{"entityId":"nebius","name":"Nebius (Token Factory)","reachPct":0.320412,"importance":0.080103,"importanceRank":26,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":1.404,"shareLowPct":0.284,"shareHighPct":6.649,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0.111,"shareLowPct":0,"shareHighPct":2.176,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"china-serving","weight":0.059,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":21,"rankP10":14,"rankP90":32,"pTop8":0,"pSelected":0.0015,"stability":"borderline"},"baselineRevenueRank":9},{"entityId":"ppio","name":"PPIO","reachPct":0.295,"importance":0.07375,"importanceRank":27,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china-serving","weight":0.059,"sharePct":5,"shareLowPct":2,"shareHighPct":12,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":27,"rankP10":19,"rankP90":34,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"ibm","name":"IBM (watsonx.ai)","reachPct":0.25419,"importance":0.0635475,"importanceRank":28,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0.37,"shareLowPct":0.093,"shareHighPct":1.101,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"independent-llm-serving","weight":0.226,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china-serving","weight":0.059,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":30,"rankP10":21,"rankP90":36,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"china-telecom-cloud","name":"China Telecom Cloud (eSurfing)","reachPct":0.236,"importance":0.059,"importanceRank":29,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china-serving","weight":0.059,"sharePct":4,"shareLowPct":1,"shareHighPct":12,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":29,"rankP10":19,"rankP90":35,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"runpod","name":"Runpod","reachPct":0.21541,"importance":0.0538525,"importanceRank":30,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":0.843,"shareLowPct":0.141,"shareHighPct":2.755,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"media-model-serving","weight":0.028,"sharePct":0.889,"shareLowPct":0,"shareHighPct":10.081,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"china-serving","weight":0.059,"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":17,"rankP90":33,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":18},{"entityId":"sambanova","name":"SambaNova (SambaCloud)","reachPct":0.193626,"importance":0.0484065,"importanceRank":31,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":0.843,"shareLowPct":0.141,"shareHighPct":3.288,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"media-model-serving","weight":0.028,"sharePct":0.111,"shareLowPct":0,"shareHighPct":2.176,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"china-serving","weight":0.059,"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":19,"rankP90":35,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"sap","name":"SAP (generative AI hub)","reachPct":0.190299,"importance":0.04757475,"importanceRank":32,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0.277,"shareLowPct":0.046,"shareHighPct":0.918,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"independent-llm-serving","weight":0.226,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china-serving","weight":0.059,"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":37,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"siliconflow","name":"SiliconFlow","reachPct":0.177,"importance":0.04425,"importanceRank":33,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china-serving","weight":0.059,"sharePct":3,"shareLowPct":1,"shareHighPct":10,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":31,"rankP10":22,"rankP90":36,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"novita-ai","name":"Novita AI","reachPct":0.139444,"importance":0.034861,"importanceRank":34,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":0.562,"shareLowPct":0.141,"shareHighPct":2.21,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"media-model-serving","weight":0.028,"sharePct":0.444,"shareLowPct":0,"shareHighPct":6.276,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"china-serving","weight":0.059,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":30,"rankP10":22,"rankP90":36,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"friendli-ai","name":"FriendliAI","reachPct":0.13012,"importance":0.03253,"importanceRank":35,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":0.562,"shareLowPct":0.085,"shareHighPct":1.667,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"media-model-serving","weight":0.028,"sharePct":0.111,"shareLowPct":0,"shareHighPct":2.176,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"china-serving","weight":0.059,"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":28,"rankP90":37,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"thinking-machines","name":"Thinking Machines (Tinker)","reachPct":0.13012,"importance":0.03253,"importanceRank":36,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":0.562,"shareLowPct":0.056,"shareHighPct":2.21,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"media-model-serving","weight":0.028,"sharePct":0.111,"shareLowPct":0,"shareHighPct":2.176,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"china-serving","weight":0.059,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":33,"rankP10":25,"rankP90":37,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"infinigence-ai","name":"Infinigence AI","reachPct":0.0885,"importance":0.022125,"importanceRank":37,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"media-model-serving","weight":0.028,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"china-serving","weight":0.059,"sharePct":1.5,"shareLowPct":0.2,"shareHighPct":6,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":36,"rankP10":30,"rankP90":37,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"chutes","name":"Chutes","reachPct":0.026544,"importance":0.006636,"importanceRank":38,"functions":[{"id":"cloud-model-platforms","weight":0.687,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"independent-llm-serving","weight":0.226,"sharePct":0.112,"shareLowPct":0.037,"shareHighPct":0.23,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"media-model-serving","weight":0.028,"sharePct":0.044,"shareLowPct":0.003,"shareHighPct":0.266,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"china-serving","weight":0.059,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":38,"rankP10":38,"rankP90":38,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null}],"selection":{"importanceIds":["microsoft-intelligent-cloud","aws","google-cloud","fireworks-ai","fal","baseten","cerebras-inference","together-ai"],"risingIds":[],"filledFromImportance":["modal","volcano-engine"]},"risingPools":{"growth":4,"shareChange":0,"committed":0},"weightSensitivity":[{"functionId":"cloud-model-platforms","end":"low","entered":[],"left":[]},{"functionId":"cloud-model-platforms","end":"high","entered":["oracle-cloud-infrastructure","databricks"],"left":["modal","volcano-engine"]},{"functionId":"independent-llm-serving","end":"low","entered":["oracle-cloud-infrastructure","databricks"],"left":["together-ai","modal"]},{"functionId":"independent-llm-serving","end":"high","entered":["groq"],"left":["volcano-engine"]},{"functionId":"media-model-serving","end":"low","entered":[],"left":[]},{"functionId":"media-model-serving","end":"high","entered":[],"left":[]},{"functionId":"china-serving","end":"low","entered":["oracle-cloud-infrastructure"],"left":["volcano-engine"]},{"functionId":"china-serving","end":"high","entered":[],"left":[]}],"validation":[],"baseline":{"spearman":0.40350877193,"baselineTop10NotSelected":["oracle-cloud-infrastructure","alibaba-cloud","databricks","snowflake","baidu-ai-cloud","nebius","digitalocean"],"selectedNotInBaselineTop10":["fireworks-ai","fal","baseten","cerebras-inference","together-ai","modal","volcano-engine"]},"raterAgreement":{"within":0.967741935484,"weightedKappa":0.651685393258,"disagreements":[]}},"inputs":{"microsoft-intelligent-cloud":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"microsoft-intelligent-cloud","name":"Microsoft Azure (Foundry)","kind":"unit","parentId":"microsoft","commercialActivityVerified":true,"shares":{"cloud-model-platforms":{"low":33.575,"central":51.312,"high":68.538,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://microsoft.com/en-us/investor/earnings/FY-2026-Q4/Document/DownloadDocument/33/TranscriptFY26Q4.docx","title":"Microsoft FY26 Fourth Quarter Earnings Conference Call (transcript)","publisher":"Microsoft","publishedDate":"2026-07-29","tier":"T5","accessedDate":"2026-10-22","locator":"Satya Nadella: 'We now have 100,000 Foundry customers, and revenue more than doubled year-over-year'; catalogue of over 11,000 models including OpenAI, Anthropic, Mistral and xAI"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$10.82bn a year (the locked card's derivation: Claude sold through cloud platforms, net of the developer's share, 3.76-4.52-5.38bn; OpenAI models sold through Azure, net, 2.4-4.8-9.6bn; other third-party and open models on platforms 0.8-1.5-3.0bn). Azure (Foundry and Azure OpenAI; GitHub Models counted here). Components of Microsoft's net serving revenue: (a) OpenAI models sold through Azure, net 2.4-4.8-9.6bn, the card's own estimate (gross US$3-12bn built from reported revenue-share payments, Microsoft taken to keep about 80%, which neither company confirms; the same sales are counted in full for OpenAI in Frontier labs, an overlap the card's known gaps record); (b) Claude through Foundry: 3-10-20% of the card's net Claude platform revenue = 0.11-0.45-1.08bn. This split is an ASSUMPTION and not disclosed by Anthropic or Microsoft; it was widened from 5-10-15% at screening because Microsoft is absent from Reuters's 2025 channel total (Anthropic's models joined Foundry in late 2025) and no source gives its present size; (c) other models (xAI, Mistral, open-weight, Meta and others): 0.15-0.30-0.80bn. Total 2.66-5.55-11.48bn. Microsoft discloses no model-serving revenue: its FY2026 Q4 earnings call (T2) says 'We now have 100,000 Foundry customers, and revenue more than doubled year-over-year', and its release says Azure revenue passed US$100bn; neither figure separates serving of other developers' models from compute and Microsoft's own products, so neither was used. Cross-checks on the platform split (none is a disclosure by the platforms): (1) Reuters on Anthropic's confidential IPO filing (T3, 30 Sep 2026): Anthropic's 2025 sales through cloud marketplaces were about US$2.16bn (47% of revenue), through Amazon and Google, with platform fees of about 16 cents per dollar; the reports give no split by platform and nothing for Microsoft. (2) The Information, as reported by PYMNTS on 18 Feb 2026 (https://www.pymnts.com/artificial-intelligence-2/2026/anthropic-set-to-pay-cloud-partners-80-billion-through-2029/, T3): Anthropic's revenue-share payments to cloud partners were about US$0.36bn in 2025 and are forecast at US$1.9bn for 2026 and US$6.4bn for 2027, with no split by partner; that forecast pre-dates Anthropic's run-rate rising past US$30bn, so it sits below the card's Claude figure. (3) SemiAnalysis (https://newsletter.semianalysis.com/p/anthropic-growth-and-bedrock-mix, T4, 27 May 2026; paywalled, the opening section was read): Bedrock is a US$5.5bn run-rate business (about 4% of AWS revenue) with 80-90% of its customers using Anthropic models, and the token-as-a-service businesses of Amazon, Microsoft and Google are together 'US$10B+ in ARR'; Anthropic is the seller of record and AWS earns an infrastructure fee plus a distribution fee, so SemiAnalysis's Bedrock figure includes infrastructure fees (counted in Infrastructure · Cloud here) and is on a different basis from this card's net figure. The three platforms' central amounts here sum to about US$10.2bn, consistent with 'US$10B+'."},"independent-llm-serving":{"participates":false,"evidence":{"url":"https://microsoft.com/en-us/investor/earnings/FY-2026-Q4/Document/DownloadDocument/33/TranscriptFY26Q4.docx","title":"Microsoft FY26 Fourth Quarter Earnings Conference Call (transcript)","publisher":"Microsoft","publishedDate":"2026-07-29","tier":"T2","accessedDate":"2026-10-22","locator":"Cloud platform: Foundry serving is counted in Cloud model platforms"}},"media-model-serving":{"participates":false,"evidence":{"url":"https://microsoft.com/en-us/investor/earnings/FY-2026-Q4/Document/DownloadDocument/33/TranscriptFY26Q4.docx","title":"Microsoft FY26 Fourth Quarter Earnings Conference Call (transcript)","publisher":"Microsoft","publishedDate":"2026-07-29","tier":"T2","accessedDate":"2026-10-22","locator":"Cloud or data platform: image, video and audio models it serves are counted in Cloud model platforms under the card scope, not in this part"}},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"baselineRevenueUsd":{"low":157224000000,"central":157224000000,"high":157224000000,"unit":"USD","basis":"Intelligent Cloud segment revenue, latest quarter × 4, US dollars","period":"Q4 FY2026 (quarter to 30 June 2026), annualised","periodEnd":"2026-06-30","source":{"url":"https://www.microsoft.com/en-us/investor/earnings/fy-2026-q4/press-release-webcast","title":"Microsoft FY26 Q4 earnings press release","publisher":"Microsoft Corporation","publishedDate":"2026-07-29","tier":"T2","accessedDate":"2026-10-22","locator":"Segment Results table: Intelligent Cloud revenue, three months ended 30 June 2026, US$39,306 million (FY2025 quarter US$29,878 million)"},"note":"Intelligent Cloud segment only (Azure and other cloud services, server products, enterprise and partner services), never Microsoft's consolidated total: US$39,306 million × 4. Full fiscal year 2026 Intelligent Cloud revenue was US$137,791 million on the same page."},"screening":{"cloud-model-platforms":{"levelLow":2,"levelHigh":4}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"08d8de1439ff6f11e156f1cc251de3d92a32e2f468729856c9a3be3bab5f1c1c"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"08d8de1439ff6f11e156f1cc251de3d92a32e2f468729856c9a3be3bab5f1c1c"}}},"aws":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"aws","name":"Amazon Web Services (AWS)","kind":"unit","parentId":"amazon","commercialActivityVerified":true,"shares":{"cloud-model-platforms":{"low":19.066,"central":29.224,"high":41.372,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.khaleejtimes.com/business/anthropic-ipo-prospectus-lays-bare-deep-dependence-on-big-tech-partners","title":"Anthropic IPO prospectus lays bare deep dependence on Big Tech partners","publisher":"Reuters (published by Khaleej Times)","publishedDate":"2026-09-30","tier":"T5","accessedDate":"2026-10-22","locator":"Anthropic's confidential IPO filing: 2025 sales through cloud marketplaces about US$2.16bn (47% of revenue), fee about 16 cents per dollar; Amazon and Google combined"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$10.82bn a year (the locked card's derivation: Claude sold through cloud platforms, net of the developer's share, 3.76-4.52-5.38bn; OpenAI models sold through Azure, net, 2.4-4.8-9.6bn; other third-party and open models on platforms 0.8-1.5-3.0bn). Amazon Bedrock (other developers' models only). Components of AWS's net serving revenue: (a) Claude through Bedrock: 40-60-80% of the card's net Claude platform revenue (3.76-4.52-5.38bn) = 1.5-2.71-4.3bn. This split is an ASSUMPTION: Reuters on the filing gives only the Amazon-plus-Google total and no split, and SemiAnalysis (below) indicates Bedrock is the largest single channel; widened from 45-60-75% at screening; (b) other models (open-weight, Mistral, OpenAI models added to Bedrock in 2026 and others): 0.30-0.45-1.10bn, part of the card's 0.8-1.5-3.0bn allowance. Total 1.80-3.16-5.40bn. Amazon's Q2 2026 release (Form 8-K exhibit, 30 Jul 2026) says Bedrock added more than 10 models including OpenAI's, Anthropic's and Google DeepMind's and that 'customers spent more in Q2 than in all prior quarters combined', but gives no Bedrock revenue; its AI run-rate of 'over $25 billion' mixes rented computing, chips and applications and was not used. Anthropic books gross marketplace sales as its own revenue and the platform's fee as a marketing cost, so the platform's retained part (about 16%) is the card's net basis. The same PYMNTS report says roughly half of Anthropic's gross profit on selling through Amazon has gone to Amazon (sources familiar with Anthropic's finances, T3), which implies Amazon keeps more than the 16% fee once infrastructure margin is counted and points to the upper half of the range above. Cross-checks on the platform split (none is a disclosure by the platforms): (1) Reuters on Anthropic's confidential IPO filing (T3, 30 Sep 2026): Anthropic's 2025 sales through cloud marketplaces were about US$2.16bn (47% of revenue), through Amazon and Google, with platform fees of about 16 cents per dollar; the reports give no split by platform and nothing for Microsoft. (2) The Information, as reported by PYMNTS on 18 Feb 2026 (https://www.pymnts.com/artificial-intelligence-2/2026/anthropic-set-to-pay-cloud-partners-80-billion-through-2029/, T3): Anthropic's revenue-share payments to cloud partners were about US$0.36bn in 2025 and are forecast at US$1.9bn for 2026 and US$6.4bn for 2027, with no split by partner; that forecast pre-dates Anthropic's run-rate rising past US$30bn, so it sits below the card's Claude figure. (3) SemiAnalysis (https://newsletter.semianalysis.com/p/anthropic-growth-and-bedrock-mix, T4, 27 May 2026; paywalled, the opening section was read): Bedrock is a US$5.5bn run-rate business (about 4% of AWS revenue) with 80-90% of its customers using Anthropic models, and the token-as-a-service businesses of Amazon, Microsoft and Google are together 'US$10B+ in ARR'; Anthropic is the seller of record and AWS earns an infrastructure fee plus a distribution fee, so SemiAnalysis's Bedrock figure includes infrastructure fees (counted in Infrastructure · Cloud here) and is on a different basis from this card's net figure. The three platforms' central amounts here sum to about US$10.2bn, consistent with 'US$10B+'."},"independent-llm-serving":{"participates":false,"evidence":{"url":"https://www.fool.com/earnings/call-transcripts/2026/08/07/amazon-amzn-q2-2026-earnings-call-transcript/","title":"Amazon (AMZN) Q2 2026 Earnings Call Transcript","publisher":"The Motley Fool (transcript of Amazon's call of 30 July 2026)","publishedDate":"2026-08-07","tier":"T2","accessedDate":"2026-10-22","locator":"Cloud platform: Bedrock serving is counted in Cloud model platforms"}},"media-model-serving":{"participates":false,"evidence":{"url":"https://www.fool.com/earnings/call-transcripts/2026/08/07/amazon-amzn-q2-2026-earnings-call-transcript/","title":"Amazon (AMZN) Q2 2026 Earnings Call Transcript","publisher":"The Motley Fool (transcript of Amazon's call of 30 July 2026)","publishedDate":"2026-08-07","tier":"T2","accessedDate":"2026-10-22","locator":"Cloud or data platform: image, video and audio models it serves are counted in Cloud model platforms under the card scope, not in this part"}},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"baselineRevenueUsd":{"low":168928000000,"central":168928000000,"high":168928000000,"unit":"USD","basis":"AWS segment net sales, latest quarter × 4, US dollars","period":"Q2 2026 (quarter to 30 June 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","tier":"T1","accessedDate":"2026-10-22","locator":"AWS segment net sales, second quarter 2026, US$42,232 million (second quarter 2025 US$30,873 million)"},"note":"AWS segment only, never Amazon's consolidated total: US$42,232 million × 4. Trailing twelve months to 30 June 2026 were US$148,404 million on the same page."},"screening":{"cloud-model-platforms":{"levelLow":2,"levelHigh":4}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"69366ab8d1eec5f72a9c131c3a2d471c2e14f1055c5ba53071d74717952ded5d"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"69366ab8d1eec5f72a9c131c3a2d471c2e14f1055c5ba53071d74717952ded5d"}}},"google-cloud":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"google-cloud","name":"Google Cloud (Vertex AI)","kind":"unit","parentId":"alphabet","commercialActivityVerified":true,"shares":{"cloud-model-platforms":{"low":7.127,"central":14.007,"high":25.716,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://futurumgroup.com/insights/alphabet-q2-fy-2026-google-cloud-leads-growth-amid-rising-ai-investment/","title":"Alphabet Q2 FY 2026: Google Cloud Leads Growth Amid Rising AI Investment","publisher":"Futurum Group (quoting Alphabet's Q2 2026 call)","publishedDate":"2026-07-23","tier":"T5","accessedDate":"2026-10-22","locator":"Alphabet's Q2 2026 call as summarised by Futurum: Cloud 'serving through Vertex and Gemini Enterprise'; backlog US$514bn; model APIs about 22 billion tokens a minute (Gemini, Google's own)"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$10.82bn a year (the locked card's derivation: Claude sold through cloud platforms, net of the developer's share, 3.76-4.52-5.38bn; OpenAI models sold through Azure, net, 2.4-4.8-9.6bn; other third-party and open models on platforms 0.8-1.5-3.0bn). Vertex AI (other developers' models only; Gemini counts for Google DeepMind in Frontier labs and its tokens were not used). Components of Google's net serving revenue: (a) Claude through Vertex: 15-28-45% of the card's net Claude platform revenue = 0.56-1.27-2.42bn. This split is an ASSUMPTION: Reuters gives only the Amazon-plus-Google total of about US$2.16bn for 2025 and no split; widened from 18-28-40% at screening; (b) other models in Model Garden (open-weight, Mistral and others): 0.15-0.25-0.80bn. Total 0.71-1.52-3.22bn. Alphabet's Q2 2026 release (T1) reports Google Cloud revenue of US$24.8bn, up 82%, and Gemini models processing 22 billion API tokens a minute (Google's own models); it does not mention Vertex AI, third-party models or Claude, so Google discloses no Vertex third-party serving revenue. Cross-checks on the platform split (none is a disclosure by the platforms): (1) Reuters on Anthropic's confidential IPO filing (T3, 30 Sep 2026): Anthropic's 2025 sales through cloud marketplaces were about US$2.16bn (47% of revenue), through Amazon and Google, with platform fees of about 16 cents per dollar; the reports give no split by platform and nothing for Microsoft. (2) The Information, as reported by PYMNTS on 18 Feb 2026 (https://www.pymnts.com/artificial-intelligence-2/2026/anthropic-set-to-pay-cloud-partners-80-billion-through-2029/, T3): Anthropic's revenue-share payments to cloud partners were about US$0.36bn in 2025 and are forecast at US$1.9bn for 2026 and US$6.4bn for 2027, with no split by partner; that forecast pre-dates Anthropic's run-rate rising past US$30bn, so it sits below the card's Claude figure. (3) SemiAnalysis (https://newsletter.semianalysis.com/p/anthropic-growth-and-bedrock-mix, T4, 27 May 2026; paywalled, the opening section was read): Bedrock is a US$5.5bn run-rate business (about 4% of AWS revenue) with 80-90% of its customers using Anthropic models, and the token-as-a-service businesses of Amazon, Microsoft and Google are together 'US$10B+ in ARR'; Anthropic is the seller of record and AWS earns an infrastructure fee plus a distribution fee, so SemiAnalysis's Bedrock figure includes infrastructure fees (counted in Infrastructure · Cloud here) and is on a different basis from this card's net figure. The three platforms' central amounts here sum to about US$10.2bn, consistent with 'US$10B+'."},"independent-llm-serving":{"participates":false,"evidence":{"url":"https://futurumgroup.com/insights/alphabet-q2-fy-2026-google-cloud-leads-growth-amid-rising-ai-investment/","title":"Alphabet Q2 FY 2026: Google Cloud Leads Growth Amid Rising AI Investment","publisher":"Futurum Group (quoting Alphabet's Q2 2026 call)","publishedDate":"2026-07-23","tier":"T3","accessedDate":"2026-10-22","locator":"Cloud platform: Vertex AI serving is counted in Cloud model platforms"}},"media-model-serving":{"participates":false,"evidence":{"url":"https://futurumgroup.com/insights/alphabet-q2-fy-2026-google-cloud-leads-growth-amid-rising-ai-investment/","title":"Alphabet Q2 FY 2026: Google Cloud Leads Growth Amid Rising AI Investment","publisher":"Futurum Group (quoting Alphabet's Q2 2026 call)","publishedDate":"2026-07-23","tier":"T3","accessedDate":"2026-10-22","locator":"Cloud or data platform: image, video and audio models it serves are counted in Cloud model platforms under the card scope, not in this part"}},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"baselineRevenueUsd":{"low":99072000000,"central":99072000000,"high":99072000000,"unit":"USD","basis":"Google Cloud segment revenue, latest quarter × 4, US dollars","period":"Q2 2026 (quarter to 30 June 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","tier":"T1","accessedDate":"2026-10-22","locator":"Google Cloud revenues, second quarter 2026, US$24,768 million (second quarter 2025 US$13,624 million), up 82%"},"note":"Google Cloud segment only, never Alphabet's consolidated total: US$24,768 million × 4. The release gives no trailing-twelve-month Cloud figure."},"screening":{"cloud-model-platforms":{"levelLow":2,"levelHigh":4}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"8e4ebfa6b3291c40ffdc1c9a532b4b617e7ae0b7e0b5010a5486968ba1484ac9"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"8e4ebfa6b3291c40ffdc1c9a532b4b617e7ae0b7e0b5010a5486968ba1484ac9"}}},"fireworks-ai":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"fireworks-ai","name":"Fireworks AI","kind":"private","commercialActivityVerified":true,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://fireworks.ai/blog/series-d-announcement","title":"Fireworks Secures $1.5 Billion in Series D Funding","publisher":"Fireworks AI","publishedDate":"2026-07-15","tier":"T2","accessedDate":"2026-10-22","locator":"Inference specialist, not a hyperscaler, enterprise-software or data platform: its managed serving is counted in Independent inference providers"}},"independent-llm-serving":{"low":27.066,"central":28.09,"high":35.354,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Annualised run-rate stated 15 July 2026","periodEnd":"2026-07-15","source":{"url":"https://fireworks.ai/blog/series-d-announcement","title":"Fireworks Secures $1.5 Billion in Series D Funding","publisher":"Fireworks AI","publishedDate":"2026-07-15","tier":"T2","accessedDate":"2026-10-22","locator":"'surpasses $1 billion in annualized revenue run rate'; 'serves more than 40 trillion tokens every day'; more than 95% from models specialised on customers' data"},"note":"Share derived from a company statement (T2). Share = company ÷ part total, with the other companies held at their central values. Part total: US$3.56bn a year (the locked card's derivation: Fireworks 1.0, Together 0.45, Baseten 0.57, Modal 0.39, Cerebras 0.25, all others 0.9; range 2.39-5.46). Revenue is counted only for managed serving of other developers' models, net of GPU rental, training and capacity a model developer buys for its own users. The company says its annualised revenue run-rate is above US$1bn (T2). Serving share 95-100% (counted as serving of customer-specialised and open models; no revenue mix is published); the high end 1.4bn allows for growth since the post. Open models carry no developer share, so gross equals net. Range is the company's figure plus the allowance for growth; no tier width added. Earlier data point: the company's Series C post (28 Oct 2025, https://fireworks.ai/blog/series-c, T2) said annualised revenue had passed US$280 million, so the run-rate rose about 3.6 times in 8.5 months. That date is 13 days short of the 9-month minimum before the 15 July 2026 figure, and no earlier primary figure was found (arr.club, tipranks and similar pages are aggregators), so there is no growth or share-change component."},"media-model-serving":{"low":0,"central":0.444,"high":6.276,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://fireworks.ai/blog/series-d-announcement","title":"Fireworks Secures $1.5 Billion in Series D Funding","publisher":"Fireworks AI","publishedDate":"2026-07-15","tier":"T5","accessedDate":"2026-10-22","locator":"Series D post (language-model tokens only)"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$0.45bn a year (the locked card's derivation: fal 0.35 net, Replicate, Runware and others 0.10; range 0.23-0.97). No media revenue split is disclosed; this is a small allowance (the company may serve none) inside the card's 0.10bn for Replicate, Runware and other hosts. "},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"baselineRevenueUsd":{"low":1000000000,"central":1000000000,"high":1000000000,"unit":"USD","basis":"Annualised revenue run-rate, all products, US dollars","period":"Annualised run-rate stated 15 July 2026","periodEnd":"2026-07-15","source":{"url":"https://fireworks.ai/blog/series-d-announcement","title":"Fireworks Secures $1.5 Billion in Series D Funding","publisher":"Fireworks AI","publishedDate":"2026-07-15","tier":"T2","accessedDate":"2026-10-22","locator":"'surpasses $1 billion in annualized revenue run rate'; 'serves more than 40 trillion tokens every day'; more than 95% from models specialised on customers' data"},"note":"The company says its annualised revenue run-rate 'surpasses $1 billion' (a stated threshold). It includes serving, fine-tuning and any training on the platform; the serving part is shown in the share above."},"screening":{"independent-llm-serving":{"levelLow":1,"levelHigh":3},"media-model-serving":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"e9100961efd624c5f62cf6ab715ff6c3bffc97ae048d5b63bcb68f53e7e2a225"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"e9100961efd624c5f62cf6ab715ff6c3bffc97ae048d5b63bcb68f53e7e2a225"}}},"fal":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"fal","name":"fal","kind":"private","commercialActivityVerified":true,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://techcrunch.com/2025/12/09/fal-nabs-140m-in-fresh-funding-led-by-sequoia-tripling-valuation-to-4-5b/","title":"Fal nabs $140M in fresh funding led by Sequoia, tripling valuation to $4.5B","publisher":"TechCrunch","publishedDate":"2025-12-09","tier":"T3","accessedDate":"2026-10-22","locator":"Independent host of image, video and audio models for developers, not a cloud or data platform"}},"independent-llm-serving":{"low":0,"central":0.014,"high":0.28,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://techcrunch.com/2025/12/09/fal-nabs-140m-in-fresh-funding-led-by-sequoia-tripling-valuation-to-4-5b/","title":"Fal nabs $140M in fresh funding led by Sequoia, tripling valuation to $4.5B","publisher":"TechCrunch","publishedDate":"2025-12-09","tier":"T5","accessedDate":"2026-10-22","locator":"Article: fal 'hosts image, video, and audio AI models'"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$3.56bn a year (the locked card's derivation: Fireworks 1.0, Together 0.45, Baseten 0.57, Modal 0.39, Cerebras 0.25, all others 0.9; range 2.39-5.46). Revenue is counted only for managed serving of other developers' models, net of GPU rental, training and capacity a model developer buys for its own users. fal serves language models only incidentally, if at all; allowance. "},"media-model-serving":{"low":70.588,"central":73.684,"high":87.805,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"February-March 2026, annualised (reported)","periodEnd":"2026-02-28","source":{"url":"https://www.techloy.com/an-ai-startups-new-funding-round-could-value-it-at-8-billion-with-two-different-price-tags/","title":"An AI startup's new funding round could value it at $8 billion with two different price tags","publisher":"Techloy (citing The Information)","publishedDate":"2026-03-19","tier":"T5","accessedDate":"2026-10-22","locator":"'annualised revenue recently reached $400 million, up from $200 million in October' (The Information, as reported by Techloy)"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$0.45bn a year (the locked card's derivation: fal 0.35 net, Replicate, Runware and others 0.10; range 0.23-0.97), so the others are held at 0.10bn. Annualised revenue (gross of payments to the developers of the proprietary models fal hosts) reached US$400m in February-March 2026: The Information (T3), as reported by Techloy on 19 March 2026, says it 'recently reached $400 million, up from $200 million in October'; Sacra (T4) estimates US$400m in February 2026. Net of payments to model developers: fal keeps 60-70-80% (the card's assumption) = 0.24-0.28-0.32bn at that figure. The high end, 0.72bn, is the card's and allows for growth since February: a later report of about US$800m appears only on aggregator pages (Digg) and is not used. Earlier figures: more than US$200m as of October 2025 (TechCrunch, 9 Dec 2025, citing Bloomberg, T3). The card's central net figure of 0.35bn was a projection; this uses the latest sourced run-rate."},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"activityGrowth":{"current":{"low":400000000,"central":400000000,"high":400000000,"unit":"USD","basis":"Annualised revenue, gross of payments to model developers, US dollars","period":"February 2026, annualised (research-firm estimate)","periodEnd":"2026-02-28","source":{"url":"https://sacra.com/c/fal-ai/","title":"fal.ai (company research page)","publisher":"Sacra","publishedDate":"undated","tier":"T4","accessedDate":"2026-10-22","locator":"'Sacra estimates that Fal.ai hit $400M in annualized revenue in February 2026'"},"note":"Sacra's estimate; The Information's figure (T3, March 2026) is the same."},"prior":{"low":25000000,"central":25000000,"high":25000000,"unit":"USD","basis":"Annualised revenue, gross of payments to model developers, US dollars","period":"End of 2024, annualised (research-firm estimate)","periodEnd":"2024-12-31","source":{"url":"https://sacra.com/c/fal-ai/","title":"fal.ai (company research page)","publisher":"Sacra","publishedDate":"undated","tier":"T4","accessedDate":"2026-10-22","locator":"Sacra's end-2024 baseline of US$25m, from which it computes growth of about 1,040% to the end of 2025"},"note":"Sacra's estimate for the end of 2024. Growth of 16 times; the method caps growth at +300%, so the exact prior does not change the score."}},"baselineRevenueUsd":{"low":400000000,"central":400000000,"high":400000000,"unit":"USD","basis":"Annualised revenue, gross of payments to model developers, US dollars","period":"February 2026, annualised (research-firm estimate)","periodEnd":"2026-02-28","source":{"url":"https://sacra.com/c/fal-ai/","title":"fal.ai (company research page)","publisher":"Sacra","publishedDate":"undated","tier":"T4","accessedDate":"2026-10-22","locator":"'Sacra estimates that Fal.ai hit $400M in annualized revenue in February 2026'"},"note":"Gross revenue, including payments passed to the developers of proprietary models. Sacra's estimate; The Information (T3) reports the same figure."},"screening":{"independent-llm-serving":{"levelLow":1,"levelHigh":2},"media-model-serving":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"00b0f26e3063142034ed5dd2db8647b68102619df881b78d664829839458b3c8"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"00b0f26e3063142034ed5dd2db8647b68102619df881b78d664829839458b3c8"}}},"baseten":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"baseten","name":"Baseten","kind":"private","commercialActivityVerified":true,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://baseten.co/blog/announcing-our-series-f/","title":"Announcing our Series F","publisher":"Baseten","publishedDate":"2026-06-22","tier":"T2","accessedDate":"2026-10-22","locator":"Inference specialist, not a hyperscaler, enterprise-software or data platform: its managed serving is counted in Independent inference providers"}},"independent-llm-serving":{"low":8.282,"central":16.011,"high":25.063,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"March 2026, annualised (research-firm estimate)","periodEnd":"2026-03-31","source":{"url":"https://sacra.com/c/baseten/","title":"Baseten (company research page)","publisher":"Sacra","publishedDate":"undated","tier":"T5","accessedDate":"2026-10-22","locator":"Sacra 'estimates that Baseten hit $600M in annualized revenue in March 2026', up from $200M in December 2025"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$3.56bn a year (the locked card's derivation: Fireworks 1.0, Together 0.45, Baseten 0.57, Modal 0.39, Cerebras 0.25, all others 0.9; range 2.39-5.46). Revenue is counted only for managed serving of other developers' models, net of GPU rental, training and capacity a model developer buys for its own users. Sacra (research firm, T4) estimates annualised revenue of US$600m in March 2026 (US$200m in December 2025); the company publishes no figure. Serving share 90-95-100% (Baseten also sells post-training, which is excluded) = 0.57bn. The range 0.27-1.0bn is the card's, and its upper end allows for growth since March: the company says (T2, Series F post, 22 Jun 2026) that 'over the last year, our revenue has grown 20x'. Revenue mix between dedicated deployments and Model APIs is not published; both are managed serving of other developers' or customers' models and count here."},"media-model-serving":{"low":0,"central":0.444,"high":6.276,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://baseten.co/blog/announcing-our-series-f/","title":"Announcing our Series F","publisher":"Baseten","publishedDate":"2026-06-22","tier":"T5","accessedDate":"2026-10-22","locator":"Series F post (no media split)"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$0.45bn a year (the locked card's derivation: fal 0.35 net, Replicate, Runware and others 0.10; range 0.23-0.97). No media revenue split is disclosed; this is a small allowance (the company may serve none) inside the card's 0.10bn for Replicate, Runware and other hosts. "},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"activityGrowth":{"current":{"low":600000000,"central":600000000,"high":600000000,"unit":"USD","basis":"Annualised revenue run-rate, all products, US dollars","period":"March 2026, annualised (research-firm estimate)","periodEnd":"2026-03-31","source":{"url":"https://sacra.com/c/baseten/","title":"Baseten (company research page)","publisher":"Sacra","publishedDate":"undated","tier":"T4","accessedDate":"2026-10-22","locator":"'Sacra estimates that Baseten hit $600M in annualized revenue in March 2026'"},"note":"Research-firm estimate. All products, so it includes post-training, which the share above excludes."},"prior":{"low":30000000,"central":30000000,"high":30000000,"unit":"USD","basis":"Annualised revenue run-rate, all products, US dollars","period":"March 2025, annualised (derived from a stated growth rate)","periodEnd":"2025-03-31","source":{"url":"https://sacra.com/c/baseten/","title":"Baseten (company research page)","publisher":"Sacra","publishedDate":"undated","tier":"T4","accessedDate":"2026-10-22","locator":"Sacra reports growth of about 1,900% year over year, i.e. about 20 times"},"note":"Derived: US$600m ÷ (1 + 19) = US$30m, using Sacra's stated year-over-year growth of about 1,900%. It agrees with the company's own statement (T2, 22 Jun 2026) that revenue 'has grown 20x' over the last year. Growth of this size is capped at +300% by the method, so the exact prior does not change the score."}},"baselineRevenueUsd":{"low":600000000,"central":600000000,"high":600000000,"unit":"USD","basis":"Annualised revenue run-rate, all products, US dollars","period":"March 2026, annualised (research-firm estimate)","periodEnd":"2026-03-31","source":{"url":"https://sacra.com/c/baseten/","title":"Baseten (company research page)","publisher":"Sacra","publishedDate":"undated","tier":"T4","accessedDate":"2026-10-22","locator":"'Sacra estimates that Baseten hit $600M in annualized revenue in March 2026'"},"note":"Research-firm estimate (T4), not a company figure."},"screening":{"independent-llm-serving":{"levelLow":1,"levelHigh":3},"media-model-serving":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"cc432e683c1833cb06e43bbbd209ad5ec5d6434946a0790ca626c0498d2afcac"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"cc432e683c1833cb06e43bbbd209ad5ec5d6434946a0790ca626c0498d2afcac"}}},"cerebras-inference":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"cerebras-inference","name":"Cerebras Inference","kind":"unit","parentId":"cerebras","commercialActivityVerified":true,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://www.storagenewsletter.com/2026/08/14/cerebras-systems-fiscal-2q26-financial-results/","title":"Cerebras Systems: fiscal 2Q26 financial results","publisher":"Cerebras Systems (results release, as published by StorageNewsletter)","publishedDate":"2026-08-14","tier":"T2","accessedDate":"2026-10-22","locator":"Inference specialist, not a hyperscaler, enterprise-software or data platform: its managed serving is counted in Independent inference providers"}},"independent-llm-serving":{"low":4.335,"central":7.022,"high":9.563,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q2 2026 (quarter to 30 June), annualised","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0002021728/000162828026056186/cbrsannouncesfinancialresu.htm","title":"Cerebras Systems Inc. announces financial results for the second quarter of 2026 (Form 8-K exhibit)","publisher":"Cerebras Systems Inc.","publishedDate":"2026-08-12","tier":"T5","accessedDate":"2026-10-22","locator":"Condensed Consolidated Statements of Operations: cloud and other services revenue US$125,991 thousand (three months ended 30 June 2026) against US$33,027 thousand (30 June 2025); hardware US$54,119 thousand; total US$180,110 thousand"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$3.56bn a year (the locked card's derivation: Fireworks 1.0, Together 0.45, Baseten 0.57, Modal 0.39, Cerebras 0.25, all others 0.9; range 2.39-5.46). Revenue is counted only for managed serving of other developers' models, net of GPU rental, training and capacity a model developer buys for its own users. Cloud serving only, never hardware sales. Cloud and other services revenue was US$125.991m in Q2 2026 (Form 8-K exhibit, T1) × 4 = 0.504bn. The filings do not split it by customer or product: the Q1 2026 10-Q (https://www.sec.gov/Archives/edgar/data/0002021728/000162828026044981/cbrs-20260331.htm) shows two unnamed related-party customers at 63% and 11% of total revenue, and says a significant part of the US$25.0bn remaining performance obligations is attributable to its master agreement with OpenAI, which is capacity sold to a model developer (it counts in Infrastructure · Cloud where it serves that developer's own users, so it is not a backlog of serving for others). The same 10-Q says the first-quarter rise in this line was 'primarily due to an increase of $49.9 million in non-pass-through related cloud and other services revenue due to increased demand for our cloud inference services', and that the line also holds support revenue on the installed hardware base and pass-through data-centre revenue (US$4.1m more than a year earlier); its inference is sold pay-as-you-go (On-Demand) or as take-or-pay Dedicated Capacity, with no revenue split. Serving share 30-50-70% = 0.15-0.25-0.35bn (the card's derivation). Cerebras's fast-inference chips are its edge, so a higher replacement level is allowed."},"media-model-serving":{"low":0,"central":0.111,"high":2.176,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.storagenewsletter.com/2026/08/14/cerebras-systems-fiscal-2q26-financial-results/","title":"Cerebras Systems: fiscal 2Q26 financial results","publisher":"Cerebras Systems (results release, as published by StorageNewsletter)","publishedDate":"2026-08-14","tier":"T5","accessedDate":"2026-10-22","locator":"Results release (no media split)"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$0.45bn a year (the locked card's derivation: fal 0.35 net, Replicate, Runware and others 0.10; range 0.23-0.97). No media revenue split is disclosed; this is a small allowance (the company may serve none) inside the card's 0.10bn for Replicate, Runware and other hosts. "},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"activityGrowth":{"current":{"low":503964000,"central":503964000,"high":503964000,"unit":"USD","basis":"Cloud and other services revenue, latest quarter × 4, US dollars","period":"Q2 2026 (quarter to 30 June 2026), annualised","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0002021728/000162828026056186/cbrsannouncesfinancialresu.htm","title":"Cerebras Systems Inc. announces financial results for the second quarter of 2026 (Form 8-K exhibit)","publisher":"Cerebras Systems Inc.","publishedDate":"2026-08-12","tier":"T1","accessedDate":"2026-10-22","locator":"Condensed Consolidated Statements of Operations: cloud and other services revenue US$125,991 thousand (three months ended 30 June 2026) against US$33,027 thousand (30 June 2025); hardware US$54,119 thousand; total US$180,110 thousand"},"note":"US$125.991m × 4. This is the whole cloud and other services line, including capacity sold to model developers for their own users, which the share above excludes; the filings do not split it."},"prior":{"low":132108000,"central":132108000,"high":132108000,"unit":"USD","basis":"Cloud and other services revenue, latest quarter × 4, US dollars","period":"Q2 2025 (quarter to 30 June 2025), annualised","periodEnd":"2025-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0002021728/000162828026056186/cbrsannouncesfinancialresu.htm","title":"Cerebras Systems Inc. announces financial results for the second quarter of 2026 (Form 8-K exhibit)","publisher":"Cerebras Systems Inc.","publishedDate":"2026-08-12","tier":"T1","accessedDate":"2026-10-22","locator":"Condensed Consolidated Statements of Operations: cloud and other services revenue US$125,991 thousand (three months ended 30 June 2026) against US$33,027 thousand (30 June 2025); hardware US$54,119 thousand; total US$180,110 thousand"},"note":"US$33.027m × 4, the prior-year column of the same statement. Growth of 3.8 times (+281% on the quarter)."}},"baselineRevenueUsd":{"low":503964000,"central":503964000,"high":503964000,"unit":"USD","basis":"Cloud and other services revenue (the cloud business as scored), latest quarter × 4, US dollars","period":"Q2 2026 (quarter to 30 June 2026), annualised","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0002021728/000162828026056186/cbrsannouncesfinancialresu.htm","title":"Cerebras Systems Inc. announces financial results for the second quarter of 2026 (Form 8-K exhibit)","publisher":"Cerebras Systems Inc.","publishedDate":"2026-08-12","tier":"T1","accessedDate":"2026-10-22","locator":"Condensed Consolidated Statements of Operations: cloud and other services revenue US$125,991 thousand (three months ended 30 June 2026) against US$33,027 thousand (30 June 2025); hardware US$54,119 thousand; total US$180,110 thousand"},"note":"Cloud and other services only, never Cerebras's hardware revenue or consolidated total (US$180.1m for the quarter)."},"screening":{"independent-llm-serving":{"levelLow":2,"levelHigh":4},"media-model-serving":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"e00a22a16dfbc29ef77895843e8a039be85a46aba3550a3fbec1a27f5389c397"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"e00a22a16dfbc29ef77895843e8a039be85a46aba3550a3fbec1a27f5389c397"}}},"together-ai":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"together-ai","name":"Together AI","kind":"private","commercialActivityVerified":true,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://techcrunch.com/2026/07/01/neocloud-together-ai-raises-800m-leaps-to-8-3b-valuation/","title":"Neocloud Together AI raises $800M, leaps to $8.3B valuation","publisher":"TechCrunch","publishedDate":"2026-07-01","tier":"T3","accessedDate":"2026-10-22","locator":"Inference specialist, not a hyperscaler, enterprise-software or data platform: its managed serving is counted in Independent inference providers"}},"independent-llm-serving":{"low":7.164,"central":12.64,"high":18.799,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Quarter to June 2026, annualised bookings","periodEnd":"2026-06-30","source":{"url":"https://techcrunch.com/2026/07/01/neocloud-together-ai-raises-800m-leaps-to-8-3b-valuation/","title":"Neocloud Together AI raises $800M, leaps to $8.3B valuation","publisher":"TechCrunch","publishedDate":"2026-07-01","tier":"T5","accessedDate":"2026-10-22","locator":"Together 'claims annual bookings of over $1.15 billion as of its last quarter'; 'thousands of paying customers'; described as an AI neocloud that rents out Nvidia GPU clusters"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$3.56bn a year (the locked card's derivation: Fireworks 1.0, Together 0.45, Baseten 0.57, Modal 0.39, Cerebras 0.25, all others 0.9; range 2.39-5.46). Revenue is counted only for managed serving of other developers' models, net of GPU rental, training and capacity a model developer buys for its own users. Bookings are not revenue and mix GPU-cluster rental with per-token serving. Revenue 0.8-1.0-1.2bn × serving share 30-45-60% = 0.24-0.45-0.72bn (the card's derivation; Sacra (T4, company research page, opened) estimates annualised revenue of about US$618m at the end of 2025 and US$1bn in February 2026, and says the per-token API business is roughly 30-40% of revenue (GPU rental is the larger share); dedicated endpoints that Together runs and manages also count here, so 30-45-60% of revenue is kept). Together is also a GPU cloud, so most of it is rental (Infrastructure · Cloud). Annual bookings of US$1.15bn (company claim, TechCrunch) are of the same order as Sacra's US$1bn revenue estimate. No in-window prior-period figure from a usable source was found, so there is no growth or share-change component."},"media-model-serving":{"low":0,"central":0.889,"high":10.081,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://techcrunch.com/2026/07/01/neocloud-together-ai-raises-800m-leaps-to-8-3b-valuation/","title":"Neocloud Together AI raises $800M, leaps to $8.3B valuation","publisher":"TechCrunch","publishedDate":"2026-07-01","tier":"T5","accessedDate":"2026-10-22","locator":"TechCrunch article (no media split)"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$0.45bn a year (the locked card's derivation: fal 0.35 net, Replicate, Runware and others 0.10; range 0.23-0.97). No media revenue split is disclosed; this is a small allowance (the company may serve none) inside the card's 0.10bn for Replicate, Runware and other hosts. "},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"baselineRevenueUsd":{"low":1000000000,"central":1000000000,"high":1000000000,"unit":"USD","basis":"Annualised revenue, all products (GPU clusters, per-token API, dedicated endpoints), US dollars","period":"February 2026, annualised (estimate)","periodEnd":"2026-02-28","source":{"url":"https://sacra.com/c/together-ai/","title":"Together AI (company research page)","publisher":"Sacra","publishedDate":"undated","tier":"T4","accessedDate":"2026-10-22","locator":"Sacra estimates annualised revenue of about US$618m at end-2025 and US$1bn in February 2026; annual bookings topped US$1.15bn in Q2 2026"},"note":"A research-firm estimate (T4), not a company figure. The company's own figure is 'annual bookings of over $1.15 billion', which is bookings and not revenue, so it is not used here."},"screening":{"independent-llm-serving":{"levelLow":1,"levelHigh":3},"media-model-serving":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"911bd56f2e535920c161458360e2f145c05c99c8c1cedd1827a5c384b500423e"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"911bd56f2e535920c161458360e2f145c05c99c8c1cedd1827a5c384b500423e"}}},"modal":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"modal","name":"Modal","kind":"private","commercialActivityVerified":true,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://modal.com/blog/1-billion-annualized-run-rate-series-d","title":"Modal blog post: $1B annualized run rate and Series D (14 October 2026)","publisher":"Modal","publishedDate":"2026-10-14","tier":"T2","accessedDate":"2026-10-22","locator":"Inference specialist, not a hyperscaler, enterprise-software or data platform: its managed serving is counted in Independent inference providers"}},"independent-llm-serving":{"low":8.116,"central":10.955,"high":13.388,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Annualised run-rate stated 14 October 2026","periodEnd":"2026-10-14","source":{"url":"https://modal.com/blog/1-billion-annualized-run-rate-series-d","title":"Modal blog post: $1B annualized run rate and Series D (14 October 2026)","publisher":"Modal","publishedDate":"2026-10-14","tier":"T5","accessedDate":"2026-10-22","locator":"'We recently crossed $1B in annualized run rate'; Sandboxes 'a $300M run rate product just on its own'"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$3.56bn a year (the locked card's derivation: Fireworks 1.0, Together 0.45, Baseten 0.57, Modal 0.39, Cerebras 0.25, all others 0.9; range 2.39-5.46). Revenue is counted only for managed serving of other developers' models, net of GPU rental, training and capacity a model developer buys for its own users. The company says its run-rate passed US$1bn, of which Sandboxes is US$300m (T2), leaving 0.7bn that mixes inference, training and batch jobs (no split published). Serving share of the remainder 40-55-70% = 0.28-0.39-0.49bn. Sandboxes are excluded. Modal's own post is the only source of the figure; TechCrunch (11 Feb 2026, T3, anonymous sources) put Modal's annualised revenue at about US$50m."},"media-model-serving":{"low":0,"central":0.889,"high":11.858,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://modal.com/blog/1-billion-annualized-run-rate-series-d","title":"Modal blog post: $1B annualized run rate and Series D (14 October 2026)","publisher":"Modal","publishedDate":"2026-10-14","tier":"T5","accessedDate":"2026-10-22","locator":"Blog post names media-generation customers (Runway, Suno); no split"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$0.45bn a year (the locked card's derivation: fal 0.35 net, Replicate, Runware and others 0.10; range 0.23-0.97). No media revenue split is disclosed; this is a small allowance (the company may serve none) inside the card's 0.10bn for Replicate, Runware and other hosts. "},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"activityGrowth":{"current":{"low":1000000000,"central":1000000000,"high":1000000000,"unit":"USD","basis":"Annualised revenue run-rate, all products (including Sandboxes), US dollars","period":"Annualised run-rate stated 14 October 2026","periodEnd":"2026-10-14","source":{"url":"https://modal.com/blog/1-billion-annualized-run-rate-series-d","title":"Series D: $1B in annualized run rate, new funding, and where Modal goes next","publisher":"Modal Labs","publishedDate":"2026-10-14","tier":"T2","accessedDate":"2026-10-22","locator":"'We recently crossed $1B in annualized run rate'; since the Series C 'usage has tripled' and Modal has 'grown more than 15x year-over-year'"},"note":"The company says it crossed US$1bn (a stated threshold). Sandboxes (US$300m) are included here and excluded from the share above."},"prior":{"low":50000000,"central":66700000,"high":100000000,"unit":"USD","basis":"Annualised revenue run-rate, all products (including Sandboxes), US dollars","period":"October 2025, annualised (derived from a stated growth multiple)","periodEnd":"2025-10-14","source":{"url":"https://modal.com/blog/1-billion-annualized-run-rate-series-d","title":"Series D: $1B in annualized run rate, new funding, and where Modal goes next","publisher":"Modal Labs","publishedDate":"2026-10-14","tier":"T5","accessedDate":"2026-10-22","locator":"'We recently crossed $1B in annualized run rate'; since the Series C 'usage has tripled' and Modal has 'grown more than 15x year-over-year'"},"note":"Derived: US$1bn ÷ 15 = US$66.7m, reading the company's 'grown more than 15x year-over-year' as run-rate growth (the post does not name the measure, so this is an interpretation). TechCrunch (11 Feb 2026, T3) put the run-rate at about US$50m, so the prior is probably no higher than this. Excluding Sandboxes entirely, current serving-related run-rate (up to US$0.7bn) is still more than ten times the prior, so the +300% cap applies under any reading."}},"baselineRevenueUsd":{"low":1000000000,"central":1000000000,"high":1000000000,"unit":"USD","basis":"Annualised revenue run-rate, all products (including Sandboxes), US dollars","period":"Annualised run-rate stated 14 October 2026","periodEnd":"2026-10-14","source":{"url":"https://modal.com/blog/1-billion-annualized-run-rate-series-d","title":"Series D: $1B in annualized run rate, new funding, and where Modal goes next","publisher":"Modal Labs","publishedDate":"2026-10-14","tier":"T2","accessedDate":"2026-10-22","locator":"'We recently crossed $1B in annualized run rate'; since the Series C 'usage has tripled' and Modal has 'grown more than 15x year-over-year'"},"note":"The company says it crossed US$1bn (a stated threshold); it includes Sandboxes (US$300m) and training and batch jobs as well as inference."},"screening":{"independent-llm-serving":{"levelLow":1,"levelHigh":3},"media-model-serving":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"238a5c994290dcfc45a844a292a8734b6424582d02d548329ecd98fceed37af6"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"238a5c994290dcfc45a844a292a8734b6424582d02d548329ecd98fceed37af6"}}},"volcano-engine":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"volcano-engine","name":"Volcano Engine (incl. BytePlus)","kind":"unit","parentId":"bytedance","commercialActivityVerified":true,"shares":{"cloud-model-platforms":{"low":0,"central":0.092,"high":0.46,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T5","accessedDate":"2026-10-22","locator":"IDC covers China only; this is an allowance for BytePlus serving of other developers' models outside mainland China"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$10.82bn a year (the locked card's derivation: Claude sold through cloud platforms, net of the developer's share, 3.76-4.52-5.38bn; OpenAI models sold through Azure, net, 2.4-4.8-9.6bn; other third-party and open models on platforms 0.8-1.5-3.0bn). BytePlus (Volcano Engine's international arm) serving third-party models outside mainland China; no figure found. Allowance 0-0.01-0.05bn. "},"independent-llm-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"Public cloud in China: its serving of other developers' models is counted in Model serving in mainland China, as the card scope states that all serving in mainland China counts there"}},"media-model-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"Public cloud in China: its serving of other developers' models is counted in Model serving in mainland China, as the card scope states that all serving in mainland China counts there"}},"china-serving":{"low":9,"central":21,"high":45,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T5","accessedDate":"2026-10-22","locator":"China public-cloud MaaS 2025H2 tracker (2025 data): first by revenue (over 40% share) and by token calls (nearly half); MaaS revenue RMB 3.07bn in 2025, about RMB 18.6bn forecast for 2026"},"note":"5LayerCake estimate (T5). Part total: US$0.92bn a year (the locked card's derivation: IDC's Q3 2026 China MaaS run-rate of about RMB 22bn × a 30% third-party share ÷ 7.15; range 0.52-1.76). Company share = its IDC share of China public-cloud MaaS revenue × the fraction of that revenue from other developers' models ÷ 0.30. Third-party models only (Seed's own Doubao models count in Frontier labs and Open-weight). Share = 0.42 (IDC: over 40% of revenue) × 15% (7-15-30%: fraction of its MaaS revenue from other developers' models, assumed because Doubao dominates its calls) ÷ 0.30 = 21% (low 0.40 × 7% ÷ 0.30, high 0.45 × 30% ÷ 0.30)."}},"screening":{"cloud-model-platforms":{"levelLow":1,"levelHigh":2},"china-serving":{"levelLow":2,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"27290cad67ebc71b398cb03102367acd8488401d20ae558281a0f175f96eed9b"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"27290cad67ebc71b398cb03102367acd8488401d20ae558281a0f175f96eed9b"}}},"oracle-cloud-infrastructure":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"oracle-cloud-infrastructure","name":"Oracle Cloud Infrastructure (OCI)","kind":"unit","parentId":"oracle","commercialActivityVerified":false,"shares":{"cloud-model-platforms":{"low":0.375,"central":1.848,"high":5.348,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://docs.oracle.com/en-us/iaas/Content/generative-ai/pretrained-models.htm","title":"Pretrained Foundational Models in Generative AI (OCI documentation)","publisher":"Oracle","publishedDate":"undated","tier":"T5","accessedDate":"2026-10-22","locator":"Chat models from Cohere, Google (Gemini 2.5), Meta (Llama 4), OpenAI (gpt-oss) and xAI (Grok 4.x), offered on-demand and on dedicated AI clusters"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$10.82bn a year (the locked card's derivation: Claude sold through cloud platforms, net of the developer's share, 3.76-4.52-5.38bn; OpenAI models sold through Azure, net, 2.4-4.8-9.6bn; other third-party and open models on platforms 0.8-1.5-3.0bn). OCI Generative AI. Oracle discloses no revenue for this service; its large AI cloud contracts are rented GPU capacity (Infrastructure · Cloud). Allowance from the card's 0.8-1.5-3.0bn 'other models on platforms' (Oracle, Databricks, Snowflake and open models on the large clouds share it): 0.04-0.20-0.60bn. Dedicated AI clusters count here only where Oracle runs and manages the model."},"independent-llm-serving":{"participates":false,"evidence":{"url":"https://docs.oracle.com/en-us/iaas/Content/generative-ai/pretrained-models.htm","title":"Pretrained Foundational Models in Generative AI (OCI documentation)","publisher":"Oracle","publishedDate":"undated","tier":"T2","accessedDate":"2026-10-22","locator":"Cloud platform: counted in Cloud model platforms"}},"media-model-serving":{"participates":false,"evidence":{"url":"https://docs.oracle.com/en-us/iaas/Content/generative-ai/pretrained-models.htm","title":"Pretrained Foundational Models in Generative AI (OCI documentation)","publisher":"Oracle","publishedDate":"undated","tier":"T2","accessedDate":"2026-10-22","locator":"Cloud or data platform: image, video and audio models it serves are counted in Cloud model platforms under the card scope, not in this part"}},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"baselineRevenueUsd":{"low":29552000000,"central":29552000000,"high":29552000000,"unit":"USD","basis":"Oracle Cloud Infrastructure revenue, latest quarter × 4, US dollars","period":"Q1 FY2027 (quarter to 31 August 2026), annualised","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","tier":"T2","accessedDate":"2026-10-22","locator":"Cloud infrastructure (IaaS) revenue Q1 FY2027 US$7,388 million (Q1 FY2026 US$3,347 million), up 121%; fiscal 2026 OCI revenue US$18,101 million"},"note":"OCI (cloud infrastructure) only, never Oracle's consolidated total or its cloud applications: US$7,388 million × 4. The release does not mention OCI Generative AI, model serving or the Cohere, Grok or Llama models."},"screening":{"cloud-model-platforms":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"51966567821c9103ae4d4931b9908392df7dfc9b3bda2f6d0c7a9213ca2497cc"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"51966567821c9103ae4d4931b9908392df7dfc9b3bda2f6d0c7a9213ca2497cc"}}},"groq":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"groq","name":"Groq","kind":"private","commercialActivityVerified":false,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://www.implicator.ai/groq-350-million-3-5-billion-valuation-after-nvidia-deal/","title":"Groq: $350 million at a $3.5 billion valuation after the Nvidia deal","publisher":"Implicator.ai","publishedDate":"2026-08-17","tier":"T3","accessedDate":"2026-10-22","locator":"Inference specialist, not a hyperscaler, enterprise-software or data platform: its managed serving is counted in Independent inference providers"}},"independent-llm-serving":{"low":1.445,"central":4.213,"high":12.788,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.implicator.ai/groq-350-million-3-5-billion-valuation-after-nvidia-deal/","title":"Groq: $350 million at a $3.5 billion valuation after the Nvidia deal","publisher":"Implicator.ai","publishedDate":"2026-08-17","tier":"T5","accessedDate":"2026-10-22","locator":"Groq continues as an independent cloud service after Nvidia's December 2025 licence; 13 data centres, 54 MW; 'more than six million developers' (self-reported); $350m raised 17 August 2026"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$3.56bn a year (the locked card's derivation: Fireworks 1.0, Together 0.45, Baseten 0.57, Modal 0.39, Cerebras 0.25, all others 0.9; range 2.39-5.46). Revenue is counted only for managed serving of other developers' models, net of GPU rental, training and capacity a model developer buys for its own users. GroqCloud serving open models on its own LPUs. No revenue is reported: the article says Groq had been 'targeting $500 million in 2025 revenue', a target not a result, and revenue near US$100m at the time of the Nvidia deal appears only in a search snippet citing Forbes, not opened. Estimate 0.05-0.15-0.50bn. Fast custom inference chips are Groq's edge, so a higher replacement level is allowed. Update at full pass: Groq's newsroom (opened 22 Oct) lists a GroqCloud 'Expanding to Meet Demand' post (16 Feb 2026), the US$650m raise (22 Jun 2026), NVIDIA Cloud Partner status (12 Aug), a US$350m Series A (17 Aug) and Meta's Muse Spark in limited preview on GroqCloud (15 Oct); the bodies of these posts were not opened, and no revenue, customer or token figure was found. commercialActivityVerified stays false because no opened source shows paying customers or revenue."},"media-model-serving":{"low":0,"central":0.222,"high":4.264,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.implicator.ai/groq-350-million-3-5-billion-valuation-after-nvidia-deal/","title":"Groq: $350 million at a $3.5 billion valuation after the Nvidia deal","publisher":"Implicator.ai","publishedDate":"2026-08-17","tier":"T5","accessedDate":"2026-10-22","locator":"Article (no media split; speech models possible)"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$0.45bn a year (the locked card's derivation: fal 0.35 net, Replicate, Runware and others 0.10; range 0.23-0.97). No media revenue split is disclosed; this is a small allowance (the company may serve none) inside the card's 0.10bn for Replicate, Runware and other hosts. "},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"screening":{"independent-llm-serving":{"levelLow":2,"levelHigh":4},"media-model-serving":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"3c2c3aac81a6c82dfca3ff4308faf83b7098c188de04df45c89b954f0d6e43e0"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"3c2c3aac81a6c82dfca3ff4308faf83b7098c188de04df45c89b954f0d6e43e0"}}},"databricks":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"databricks","name":"Databricks","kind":"private","commercialActivityVerified":true,"shares":{"cloud-model-platforms":{"low":0.373,"central":1.386,"high":3.613,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q4 FY2026 run-rate (company), serving part estimated","periodEnd":"2026-01-31","source":{"url":"https://databricks.com/company/newsroom/press-releases/databricks-grows-65-yoy-surpasses-5-4-billion-revenue-run-rate","title":"Databricks Grows >65% YoY, Surpasses $5.4 Billion Revenue Run-Rate, Doubles Down on Lakebase and Genie","publisher":"Databricks","publishedDate":"2026-02-09","tier":"T5","accessedDate":"2026-10-22","locator":"'Crossing $1.4 billion revenue run-rate for its AI products' (9 February 2026); the release does not define the figure"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$10.82bn a year (the locked card's derivation: Claude sold through cloud platforms, net of the developer's share, 3.76-4.52-5.38bn; OpenAI models sold through Azure, net, 2.4-4.8-9.6bn; other third-party and open models on platforms 0.8-1.5-3.0bn). Mosaic AI Model Serving and Foundation Model APIs for Claude, OpenAI, Gemini and open models. The company says its AI products pass a US$1.4bn run-rate (T2); the release lists no model-serving figure, and the later release of 13 August 2026 (https://www.databricks.com/company/newsroom/press-releases/databricks-grows-80-yoy-surpasses-7b-revenue-run-rate-scales) gives none either. Serving of other developers' models, net of what is paid to them: 1.4bn × 10-25-40% model serving × 30-43-70% retained = 0.04-0.15-0.40bn. Period ends 2026-01-31 (the run-rate quarter), not stale."},"independent-llm-serving":{"participates":false,"evidence":{"url":"https://databricks.com/company/newsroom/press-releases/databricks-grows-65-yoy-surpasses-5-4-billion-revenue-run-rate","title":"Databricks Grows >65% YoY, Surpasses $5.4 Billion Revenue Run-Rate, Doubles Down on Lakebase and Genie","publisher":"Databricks","publishedDate":"2026-02-09","tier":"T2","accessedDate":"2026-10-22","locator":"Data platform: model serving is counted in Cloud model platforms"}},"media-model-serving":{"participates":false,"evidence":{"url":"https://databricks.com/company/newsroom/press-releases/databricks-grows-65-yoy-surpasses-5-4-billion-revenue-run-rate","title":"Databricks Grows >65% YoY, Surpasses $5.4 Billion Revenue Run-Rate, Doubles Down on Lakebase and Genie","publisher":"Databricks","publishedDate":"2026-02-09","tier":"T2","accessedDate":"2026-10-22","locator":"Cloud or data platform: image, video and audio models it serves are counted in Cloud model platforms under the card scope, not in this part"}},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"baselineRevenueUsd":{"low":7000000000,"central":7000000000,"high":7000000000,"unit":"USD","basis":"Total revenue run-rate, US dollars","period":"Q2 FY2027 (quarter to 31 July 2026), run-rate","periodEnd":"2026-07-31","source":{"url":"https://www.databricks.com/company/newsroom/press-releases/databricks-grows-80-yoy-surpasses-7b-revenue-run-rate-scales","title":"Databricks Grows >80% YoY, Surpasses $7B Revenue Run-Rate, Scales ...","publisher":"Databricks","publishedDate":"2026-08-13","tier":"T2","accessedDate":"2026-10-22","locator":"'crossed a $7 billion revenue run-rate, delivering >80% year-over-year growth during its Q2'"},"note":"The company says it crossed US$7bn (a stated threshold, so the true run-rate may be a little higher). Databricks is scored as a whole company. The release gives run-rates for Lakehouse (US$1.5bn) and Lakebase (US$100m) but none for model serving or AI products as a category."},"screening":{"cloud-model-platforms":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"af3b35fdad930f51596ea3cd20b2dd7297e69c91c8fcaa77ab7dfcebe38fdebf"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"af3b35fdad930f51596ea3cd20b2dd7297e69c91c8fcaa77ab7dfcebe38fdebf"}}},"snowflake":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"snowflake","name":"Snowflake","kind":"public","commercialActivityVerified":true,"shares":{"cloud-model-platforms":{"low":0.28,"central":1.109,"high":2.727,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.saastr.com/5-interesting-learnings-from-snowflake-at-6-billion-in-revenue-37-growth-and-accelerating-126-nrr-and-a-gross-margin-guided-down-to-pay-for-ai/","title":"5 interesting learnings from Snowflake at $6 billion in revenue, 37% growth and accelerating, 126% NRR, and a gross margin guided down to pay for AI","publisher":"SaaStr","publishedDate":"undated","tier":"T5","accessedDate":"2026-10-22","locator":"AI products (CoCo, CoWork, AI functions, Cortex AI Gateway) drove roughly half of the growth acceleration; 'Snowflake buys the inference, so Snowflake pays for it'; product gross margin 74.7%"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$10.82bn a year (the locked card's derivation: Claude sold through cloud platforms, net of the developer's share, 3.76-4.52-5.38bn; OpenAI models sold through Azure, net, 2.4-4.8-9.6bn; other third-party and open models on platforms 0.8-1.5-3.0bn). Cortex AI serving Claude, OpenAI, Llama and Mistral models. Snowflake gives no AI revenue figure in the sources opened (search-result excerpts of its December 2025 call mention an AI figure of about US$100m, cut off mid-figure; that call was not opened, so the figure is not used). Net retained revenue after paying model developers, allowance within the card's 'other models on platforms': 0.03-0.12-0.30bn."},"independent-llm-serving":{"participates":false,"evidence":{"url":"https://www.saastr.com/5-interesting-learnings-from-snowflake-at-6-billion-in-revenue-37-growth-and-accelerating-126-nrr-and-a-gross-margin-guided-down-to-pay-for-ai/","title":"5 interesting learnings from Snowflake at $6 billion in revenue, 37% growth and accelerating, 126% NRR, and a gross margin guided down to pay for AI","publisher":"SaaStr","publishedDate":"undated","tier":"T3","accessedDate":"2026-10-22","locator":"Data platform: model serving is counted in Cloud model platforms"}},"media-model-serving":{"participates":false,"evidence":{"url":"https://www.saastr.com/5-interesting-learnings-from-snowflake-at-6-billion-in-revenue-37-growth-and-accelerating-126-nrr-and-a-gross-margin-guided-down-to-pay-for-ai/","title":"5 interesting learnings from Snowflake at $6 billion in revenue, 37% growth and accelerating, 126% NRR, and a gross margin guided down to pay for AI","publisher":"SaaStr","publishedDate":"undated","tier":"T3","accessedDate":"2026-10-22","locator":"Cloud or data platform: image, video and audio models it serves are counted in Cloud model platforms under the card scope, not in this part"}},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"baselineRevenueUsd":{"low":6200000000,"central":6200000000,"high":6200000000,"unit":"USD","basis":"Total revenue, latest quarter × 4, US dollars","period":"Q2 FY2027 (quarter to 31 July 2026), annualised","periodEnd":"2026-07-31","source":{"url":"https://www.01net.it/snowflake-reports-financial-results-for-the-second-quarter-of-fiscal-2027/","title":"Snowflake Reports Financial Results for the Second Quarter of Fiscal 2027","publisher":"Snowflake Inc. (release as republished by 01net)","publishedDate":"2026-09-03","tier":"T2","accessedDate":"2026-10-22","locator":"Total revenue US$1.55 billion, up 35%; product revenue US$1,491.9 million, up 37% (quarter ended 31 July 2026; release dated 2 September 2026)"},"note":"US$1.55bn × 4; the release states total revenue only to the nearest US$10 million in the headline, so the annualised figure is approximate. Snowflake is scored as a whole company. The release gives no AI revenue run-rate."},"screening":{"cloud-model-platforms":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"36f1e91ba683c1b57ea6f9c8ab2220e577377e236793debc6cb4e6e2e0125dd2"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"36f1e91ba683c1b57ea6f9c8ab2220e577377e236793debc6cb4e6e2e0125dd2"}}},"alibaba-cloud":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"alibaba-cloud","name":"Alibaba Cloud (Model Studio)","kind":"unit","parentId":"alibaba","commercialActivityVerified":true,"shares":{"cloud-model-platforms":{"low":0,"central":0.185,"high":0.735,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T5","accessedDate":"2026-10-22","locator":"IDC covers China only; this is an allowance for Model Studio serving of other developers' models outside mainland China"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$10.82bn a year (the locked card's derivation: Claude sold through cloud platforms, net of the developer's share, 3.76-4.52-5.38bn; OpenAI models sold through Azure, net, 2.4-4.8-9.6bn; other third-party and open models on platforms 0.8-1.5-3.0bn). Alibaba Cloud international serving of third-party models outside mainland China; no figure found. Allowance 0-0.02-0.08bn. "},"independent-llm-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"Public cloud in China: its serving of other developers' models is counted in Model serving in mainland China, as the card scope states that all serving in mainland China counts there"}},"media-model-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"Public cloud in China: its serving of other developers' models is counted in Model serving in mainland China, as the card scope states that all serving in mainland China counts there"}},"china-serving":{"low":5,"central":12,"high":29,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T5","accessedDate":"2026-10-22","locator":"China public-cloud MaaS 2025H2 tracker (2025 data): second by revenue and by token calls; MaaS revenue RMB 3.07bn in 2025, about RMB 18.6bn forecast for 2026"},"note":"5LayerCake estimate (T5). Part total: US$0.92bn a year (the locked card's derivation: IDC's Q3 2026 China MaaS run-rate of about RMB 22bn × a 30% third-party share ÷ 7.15; range 0.52-1.76). Company share = its IDC share of China public-cloud MaaS revenue × the fraction of that revenue from other developers' models ÷ 0.30. Third-party models only (Qwen counts in Frontier labs and Open-weight). Share = 18% (15-18-25%: IDC rank two by revenue, size assumed) × 20% (10-20-35%: fraction from other developers' models) ÷ 0.30 = 12%."}},"baselineRevenueUsd":{"low":28556000000,"central":28556000000,"high":28556000000,"unit":"USD","basis":"AI Cloud and Compute Services segment revenue, latest quarter × 4, US dollars","period":"June quarter 2026 (quarter to 30 June 2026), annualised","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0001577552/000110465926099220/tm2623667d1_ex99-1.htm","title":"Alibaba Group announces June quarter 2026 results (Form 6-K exhibit)","publisher":"Alibaba Group Holding Limited","publishedDate":"2026-08-20","tier":"T1","accessedDate":"2026-10-22","locator":"AI Cloud and Compute Services segment revenue RMB48,437 million (US$7,139 million at RMB6.7851 per US$), up 45%; AI-related product revenue RMB12,376 million, 'twelfth consecutive quarter of triple-digit' growth"},"note":"The segment that now combines Alibaba Cloud Intelligence Group with T-Head (the prior year was recast), so it is slightly wider than Alibaba Cloud alone: US$7,139 million × 4. Converted at RMB6.7851 per US$ (30 June 2026, Federal Reserve H.10, as used in the release)."},"screening":{"cloud-model-platforms":{"levelLow":1,"levelHigh":2},"china-serving":{"levelLow":2,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"b9f910bd445a2a39d9db8a3fe8801c085f4b19a8aa6f3cc2b2c8d34dbb3cb533"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"b9f910bd445a2a39d9db8a3fe8801c085f4b19a8aa6f3cc2b2c8d34dbb3cb533"}}},"deepinfra":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"deepinfra","name":"DeepInfra","kind":"private","commercialActivityVerified":true,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://www.storagenewsletter.com/?p=305786","title":"DeepInfra Surpasses $100M ARR, Triples Token Volume","publisher":"DeepInfra (press release, as published by StorageNewsletter)","publishedDate":"2026-10-05","tier":"T2","accessedDate":"2026-10-22","locator":"Inference specialist, not a hyperscaler, enterprise-software or data platform: its managed serving is counted in Independent inference providers"}},"independent-llm-serving":{"low":1.42,"central":2.528,"high":2.937,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Annualised run-rate stated 5 October 2026","periodEnd":"2026-10-05","source":{"url":"https://www.storagenewsletter.com/?p=305786","title":"DeepInfra Surpasses $100M ARR, Triples Token Volume","publisher":"DeepInfra (press release, as published by StorageNewsletter)","publishedDate":"2026-10-05","tier":"T5","accessedDate":"2026-10-22","locator":"'surpassed $100 million in ARR', 'over $105 million annualized run-rate', 22 trillion tokens a week, 200+ open-source models"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$3.56bn a year (the locked card's derivation: Fireworks 1.0, Together 0.45, Baseten 0.57, Modal 0.39, Cerebras 0.25, all others 0.9; range 2.39-5.46). Revenue is counted only for managed serving of other developers' models, net of GPU rental, training and capacity a model developer buys for its own users. The company says ARR is above US$100m and the annualised run-rate above US$105m (T2). The release also mentions dedicated AI compute and enterprise 'DeepCluster' offerings, which may be rental; serving share 50-90-100% = 0.05-0.09-0.105bn."},"media-model-serving":{"low":0,"central":0.444,"high":6.276,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.storagenewsletter.com/?p=305786","title":"DeepInfra Surpasses $100M ARR, Triples Token Volume","publisher":"DeepInfra (press release, as published by StorageNewsletter)","publishedDate":"2026-10-05","tier":"T5","accessedDate":"2026-10-22","locator":"Press release (no media split)"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$0.45bn a year (the locked card's derivation: fal 0.35 net, Replicate, Runware and others 0.10; range 0.23-0.97). No media revenue split is disclosed; this is a small allowance (the company may serve none) inside the card's 0.10bn for Replicate, Runware and other hosts. "},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"baselineRevenueUsd":{"low":105000000,"central":105000000,"high":105000000,"unit":"USD","basis":"Annualised revenue run-rate, all products, US dollars","period":"Annualised run-rate stated 5 October 2026","periodEnd":"2026-10-05","source":{"url":"https://www.storagenewsletter.com/?p=305786","title":"DeepInfra Surpasses $100M ARR, Triples Token Volume","publisher":"DeepInfra (press release, as published by StorageNewsletter)","publishedDate":"2026-10-05","tier":"T2","accessedDate":"2026-10-22","locator":"'surpassed $100 million in ARR'; annualised run-rate above US$105 million (company press release, 5 October 2026)"},"note":"The company says annualised run-rate is above US$105m (a stated threshold); it includes dedicated compute and DeepCluster offerings as well as serverless inference."},"screening":{"independent-llm-serving":{"levelLow":1,"levelHigh":2},"media-model-serving":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"a72cf95f92b531f9e6b2d796555ee32511263f2a94feac03e4a17c298f3ed1ba"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"a72cf95f92b531f9e6b2d796555ee32511263f2a94feac03e4a17c298f3ed1ba"}}},"digitalocean":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"digitalocean","name":"DigitalOcean","kind":"public","commercialActivityVerified":true,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://www.01net.it/digitalocean-announces-second-quarter-2026-financial-results/","title":"DigitalOcean Announces Second Quarter 2026 Financial Results","publisher":"DigitalOcean (press release, as reprinted by 01net.it)","publishedDate":"2026-08-04","tier":"T2","accessedDate":"2026-10-22","locator":"Counted in Independent inference providers, as in the card's weight derivation (its allowance names Nebius Token Factory, Crusoe and Cloudflare Workers AI there); the card scope text could also place a neocloud in Cloud model platforms, but each part's weight is proportional to its dollar total, so Reach is almost the same either way"}},"independent-llm-serving":{"low":0.855,"central":2.247,"high":4.396,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q2 2026 (quarter to 30 June), annualised","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0001582961/000162828026052135/a2026-q2dopressrelease.htm","title":"DigitalOcean Announces Second Quarter 2026 Financial Results (Form 8-K)","publisher":"DigitalOcean Holdings, Inc.","publishedDate":"2026-08-04","tier":"T5","accessedDate":"2026-10-22","locator":"'AI Customer ARR was $234 million, an increase of 212%'; '85% of our AI customer ARR now comes from inference and core cloud rather than bare metal'"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$3.56bn a year (the locked card's derivation: Fireworks 1.0, Together 0.45, Baseten 0.57, Modal 0.39, Cerebras 0.25, all others 0.9; range 2.39-5.46). Revenue is counted only for managed serving of other developers' models, net of GPU rental, training and capacity a model developer buys for its own users. Gradient AI / Inference Engine serverless and dedicated inference of open and closed models. On the call the chief executive said Inference Services (all non-bare-metal inferencing) are over 70% of AI customer ARR, about US$164m or more (said on the call; the transcript is on a finance portal, so it is not cited; the release does not give the figure). Only managed serving of other developers' models counts, not customers' own workloads on rented GPUs: 20-50-97% of 0.164bn = 0.03-0.08-0.16bn."},"media-model-serving":{"low":0,"central":0.111,"high":2.176,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.01net.it/digitalocean-announces-second-quarter-2026-financial-results/","title":"DigitalOcean Announces Second Quarter 2026 Financial Results","publisher":"DigitalOcean (press release, as reprinted by 01net.it)","publishedDate":"2026-08-04","tier":"T5","accessedDate":"2026-10-22","locator":"Release (no media split)"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$0.45bn a year (the locked card's derivation: fal 0.35 net, Replicate, Runware and others 0.10; range 0.23-0.97). No media revenue split is disclosed; this is a small allowance (the company may serve none) inside the card's 0.10bn for Replicate, Runware and other hosts. "},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"activityGrowth":{"current":{"low":234000000,"central":234000000,"high":234000000,"unit":"USD","basis":"AI customer ARR (quarterly AI customer revenue × 4), US dollars","period":"Q2 2026 (quarter to 30 June 2026), annualised","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0001582961/000162828026052135/a2026-q2dopressrelease.htm","title":"DigitalOcean Announces Second Quarter 2026 Financial Results (Form 8-K)","publisher":"DigitalOcean Holdings, Inc.","publishedDate":"2026-08-04","tier":"T1","accessedDate":"2026-10-22","locator":"'AI Customer ARR was $234 million, an increase of 212%'; '85% of our AI customer ARR now comes from inference and core cloud rather than bare metal'"},"note":"The release defines AI customer ARR as quarterly AI customer revenue × 4. It includes bare-metal GPU rental (15%), which the share above excludes."},"prior":{"low":75000000,"central":75000000,"high":75000000,"unit":"USD","basis":"AI customer ARR (quarterly AI customer revenue × 4), US dollars","period":"Q2 2025 (quarter to 30 June 2025), annualised","periodEnd":"2025-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0001582961/000162828026052135/a2026-q2dopressrelease.htm","title":"DigitalOcean Announces Second Quarter 2026 Financial Results (Form 8-K)","publisher":"DigitalOcean Holdings, Inc.","publishedDate":"2026-08-04","tier":"T1","accessedDate":"2026-10-22","locator":"'AI Customer ARR was $234 million, an increase of 212%'; '85% of our AI customer ARR now comes from inference and core cloud rather than bare metal'"},"note":"Derived from the same release: US$234m ÷ (1 + 2.12) = US$75.0m, using the stated increase of 212%."}},"baselineRevenueUsd":{"low":1124800000,"central":1124800000,"high":1124800000,"unit":"USD","basis":"Total revenue, latest quarter × 4, US dollars","period":"Q2 2026 (quarter to 30 June 2026), annualised","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0001582961/000162828026052135/a2026-q2dopressrelease.htm","title":"DigitalOcean Announces Second Quarter 2026 Financial Results (Form 8-K)","publisher":"DigitalOcean Holdings, Inc.","publishedDate":"2026-08-04","tier":"T1","accessedDate":"2026-10-22","locator":"'AI Customer ARR was $234 million, an increase of 212%'; '85% of our AI customer ARR now comes from inference and core cloud rather than bare metal'"},"note":"US$281.2m × 4; the release states total ARR of US$1,125m. DigitalOcean is scored as a whole company."},"screening":{"independent-llm-serving":{"levelLow":1,"levelHigh":2},"media-model-serving":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"13927914a45996982ffce64594a9fa2f86362beaafb007c0ca62a7c299e785c5"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"13927914a45996982ffce64594a9fa2f86362beaafb007c0ca62a7c299e785c5"}}},"baidu-ai-cloud":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"baidu-ai-cloud","name":"Baidu AI Cloud (Qianfan)","kind":"unit","parentId":"baidu","commercialActivityVerified":true,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"Public cloud in China: its serving of other developers' models is counted in Model serving in mainland China, as the card scope states that all serving in mainland China counts there"}},"independent-llm-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"Public cloud in China: its serving of other developers' models is counted in Model serving in mainland China, as the card scope states that all serving in mainland China counts there"}},"media-model-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"Public cloud in China: its serving of other developers' models is counted in Model serving in mainland China, as the card scope states that all serving in mainland China counts there"}},"china-serving":{"low":2,"central":8,"high":19,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T5","accessedDate":"2026-10-22","locator":"China public-cloud MaaS 2025H2 tracker (2025 data): third by revenue and by token calls; MaaS revenue RMB 3.07bn in 2025, about RMB 18.6bn forecast for 2026"},"note":"5LayerCake estimate (T5). Part total: US$0.92bn a year (the locked card's derivation: IDC's Q3 2026 China MaaS run-rate of about RMB 22bn × a 30% third-party share ÷ 7.15; range 0.52-1.76). Company share = its IDC share of China public-cloud MaaS revenue × the fraction of that revenue from other developers' models ÷ 0.30. Third-party models only (ERNIE counts in Frontier labs). Share = 10% (6-10-14%: IDC rank three by revenue, size assumed) × 25% (10-25-40%: Qianfan also serves DeepSeek, Qwen and others) ÷ 0.30 = 8%."}},"baselineRevenueUsd":{"low":4303600000,"central":4303600000,"high":4303600000,"unit":"USD","basis":"AI Cloud Infra revenue, latest quarter × 4, US dollars","period":"Q2 2026 (quarter to 30 June 2026), annualised","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0001329099/000119312526355431/d112854dex991.htm","title":"Baidu announces second quarter 2026 results (Form 6-K exhibit)","publisher":"Baidu, Inc.","publishedDate":"2026-08-18","tier":"T1","accessedDate":"2026-10-22","locator":"AI Cloud Infra revenue RMB7.3 billion in Q2 2026 (RMB4.9 billion in Q2 2025), up 50%; RMB6.7851 per US$ (30 June 2026)"},"note":"RMB7.3bn ÷ 6.7851 × 4 = US$4.30bn. AI Cloud Infra is Baidu's cloud segment (including GPU cloud, which grew 283%), reported to one decimal in RMB billions; the release does not mention Qianfan, model-service revenue or token calls."},"screening":{"china-serving":{"levelLow":2,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"83be1ec2b42b00d179483dd5a1a8e246fdfa11ac8fa063b6f61ee2de74075ec5"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"83be1ec2b42b00d179483dd5a1a8e246fdfa11ac8fa063b6f61ee2de74075ec5"}}},"china-mobile-cloud":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"china-mobile-cloud","name":"China Mobile Cloud","kind":"unit","parentId":"china-mobile","commercialActivityVerified":true,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"Public cloud in China: its serving of other developers' models is counted in Model serving in mainland China, as the card scope states that all serving in mainland China counts there"}},"independent-llm-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"Public cloud in China: its serving of other developers' models is counted in Model serving in mainland China, as the card scope states that all serving in mainland China counts there"}},"media-model-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"Public cloud in China: its serving of other developers' models is counted in Model serving in mainland China, as the card scope states that all serving in mainland China counts there"}},"china-serving":{"low":3,"central":8,"high":21,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T5","accessedDate":"2026-10-22","locator":"China public-cloud MaaS 2025H2 tracker (2025 data): fifth by revenue and by token calls; MaaS revenue RMB 3.07bn in 2025, about RMB 18.6bn forecast for 2026"},"note":"5LayerCake estimate (T5). Part total: US$0.92bn a year (the locked card's derivation: IDC's Q3 2026 China MaaS run-rate of about RMB 22bn × a 30% third-party share ÷ 7.15; range 0.52-1.76). Company share = its IDC share of China public-cloud MaaS revenue × the fraction of that revenue from other developers' models ÷ 0.30. Third-party models only. Share = 5% (4-5-8%: IDC rank five by revenue, size assumed) × 50% (25-50-80%: fraction from other developers' models) ÷ 0.30 = 8%."}},"screening":{"china-serving":{"levelLow":2,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"c3259ce9433976aad277c806bb7c0d2929d41b2c711dbd23d8b560c2bac717de"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"c3259ce9433976aad277c806bb7c0d2929d41b2c711dbd23d8b560c2bac717de"}}},"huawei-cloud":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"huawei-cloud","name":"Huawei Cloud","kind":"unit","parentId":"huawei","commercialActivityVerified":true,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"Public cloud in China: its serving of other developers' models is counted in Model serving in mainland China, as the card scope states that all serving in mainland China counts there"}},"independent-llm-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"Public cloud in China: its serving of other developers' models is counted in Model serving in mainland China, as the card scope states that all serving in mainland China counts there"}},"media-model-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"Public cloud in China: its serving of other developers' models is counted in Model serving in mainland China, as the card scope states that all serving in mainland China counts there"}},"china-serving":{"low":2,"central":7,"high":16,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T5","accessedDate":"2026-10-22","locator":"China public-cloud MaaS 2025H2 tracker (2025 data): named among public-cloud vendors, no rank given; MaaS revenue RMB 3.07bn in 2025, about RMB 18.6bn forecast for 2026"},"note":"5LayerCake estimate (T5). Part total: US$0.92bn a year (the locked card's derivation: IDC's Q3 2026 China MaaS run-rate of about RMB 22bn × a 30% third-party share ÷ 7.15; range 0.52-1.76). Company share = its IDC share of China public-cloud MaaS revenue × the fraction of that revenue from other developers' models ÷ 0.30. Third-party models only (Pangu counts elsewhere). Share = 4% (2-4-6%: named, outside the top five by revenue, size assumed) × 50% (30-50-80%: fraction from other developers' models, e.g. DeepSeek on Ascend chips) ÷ 0.30 = 7%."}},"screening":{"china-serving":{"levelLow":2,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"54eb4f56d028895228dda4e28469efe5c2e192e8a601ee275928be33b5d1a7aa"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"54eb4f56d028895228dda4e28469efe5c2e192e8a601ee275928be33b5d1a7aa"}}},"tencent-cloud":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"tencent-cloud","name":"Tencent Cloud","kind":"unit","parentId":"tencent","commercialActivityVerified":true,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"Public cloud in China: its serving of other developers' models is counted in Model serving in mainland China, as the card scope states that all serving in mainland China counts there"}},"independent-llm-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"Public cloud in China: its serving of other developers' models is counted in Model serving in mainland China, as the card scope states that all serving in mainland China counts there"}},"media-model-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"Public cloud in China: its serving of other developers' models is counted in Model serving in mainland China, as the card scope states that all serving in mainland China counts there"}},"china-serving":{"low":2,"central":7,"high":16,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T5","accessedDate":"2026-10-22","locator":"China public-cloud MaaS 2025H2 tracker (2025 data): named among public-cloud vendors, no rank given; MaaS revenue RMB 3.07bn in 2025, about RMB 18.6bn forecast for 2026"},"note":"5LayerCake estimate (T5). Part total: US$0.92bn a year (the locked card's derivation: IDC's Q3 2026 China MaaS run-rate of about RMB 22bn × a 30% third-party share ÷ 7.15; range 0.52-1.76). Company share = its IDC share of China public-cloud MaaS revenue × the fraction of that revenue from other developers' models ÷ 0.30. Third-party models only (Hunyuan counts elsewhere). Share = 4% (2-4-6%: named, outside the top five by revenue, size assumed) × 50% (30-50-80%: fraction from other developers' models, e.g. DeepSeek) ÷ 0.30 = 7%."}},"screening":{"china-serving":{"levelLow":2,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"4226399b6141543feef3f42d561849b8e5adcd0c011a0341b2d376cb969417e5"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"4226399b6141543feef3f42d561849b8e5adcd0c011a0341b2d376cb969417e5"}}},"coreweave":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"coreweave","name":"CoreWeave (W&B Inference)","kind":"public","commercialActivityVerified":true,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://www.constellationr.com/insights/news/coreweave-q2-stronger-expected-touts-backlog","title":"CoreWeave Q2 stronger than expected, touts backlog","publisher":"Constellation Research (quoting CoreWeave's Q2 2026 call)","publishedDate":"2026-08-11","tier":"T2","accessedDate":"2026-10-22","locator":"Counted in Independent inference providers, as in the card's weight derivation (its allowance names Nebius Token Factory, Crusoe and Cloudflare Workers AI there); the card scope text could also place a neocloud in Cloud model platforms, but each part's weight is proportional to its dollar total, so Reach is almost the same either way"}},"independent-llm-serving":{"low":1.13,"central":1.685,"high":2.778,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q2 2026 booked ARR","periodEnd":"2026-06-30","source":{"url":"https://www.constellationr.com/insights/news/coreweave-q2-stronger-expected-touts-backlog","title":"CoreWeave Q2 stronger than expected, touts backlog","publisher":"Constellation Research (quoting CoreWeave's Q2 2026 call)","publishedDate":"2026-08-11","tier":"T5","accessedDate":"2026-10-22","locator":"CEO Mike Intrator: 'booked ARR for our managed inference platform has grown from $1 million to more than $100 million'; 'We expect to exit 2026 with at least $250 million'"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$3.56bn a year (the locked card's derivation: Fireworks 1.0, Together 0.45, Baseten 0.57, Modal 0.39, Cerebras 0.25, all others 0.9; range 2.39-5.46). Revenue is counted only for managed serving of other developers' models, net of GPU rental, training and capacity a model developer buys for its own users. W&B Inference and other managed inference, never GPU rental. The company says booked ARR for managed inference passed US$100m (T2). Booked ARR is contracted, not recognised revenue, and part may be dedicated capacity bought by developers for their own users (Infrastructure · Cloud): counted 40-60-100% of 0.1bn = 0.04-0.06-0.10bn. CoreWeave's results release (https://investors.coreweave.com/news/news-details/2026/CoreWeave-Reports-Strong-Second-Quarter-2026-Results/default.aspx) does not mention managed inference."},"media-model-serving":{"low":0,"central":0.111,"high":2.176,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.constellationr.com/insights/news/coreweave-q2-stronger-expected-touts-backlog","title":"CoreWeave Q2 stronger than expected, touts backlog","publisher":"Constellation Research (quoting CoreWeave's Q2 2026 call)","publishedDate":"2026-08-11","tier":"T5","accessedDate":"2026-10-22","locator":"Article (no media split)"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$0.45bn a year (the locked card's derivation: fal 0.35 net, Replicate, Runware and others 0.10; range 0.23-0.97). No media revenue split is disclosed; this is a small allowance (the company may serve none) inside the card's 0.10bn for Replicate, Runware and other hosts. "},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"screening":{"independent-llm-serving":{"levelLow":1,"levelHigh":3},"media-model-serving":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"77c1f1fca429fcfca9210fb605c02613843ea0d0aa22e08a30769dfe1a2c9068"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"77c1f1fca429fcfca9210fb605c02613843ea0d0aa22e08a30769dfe1a2c9068"}}},"crusoe":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"crusoe","name":"Crusoe (Managed Inference)","kind":"private","commercialActivityVerified":true,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://siliconangle.com/2026/09/17/ai-data-center-builder-crusoe-valued-at-30-9b-in-3-9b-round/","title":"AI data center builder Crusoe valued at $30.9B in $3.9B round","publisher":"SiliconANGLE (quoting Crusoe)","publishedDate":"2026-09-17","tier":"T2","accessedDate":"2026-10-22","locator":"Counted in Independent inference providers, as in the card's weight derivation (its allowance names Nebius Token Factory, Crusoe and Cloudflare Workers AI there); the card scope text could also place a neocloud in Cloud model platforms, but each part's weight is proportional to its dollar total, so Reach is almost the same either way"}},"independent-llm-serving":{"low":0.85,"central":1.685,"high":2.778,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Annualised, stated 17 September 2026","periodEnd":"2026-09-17","source":{"url":"https://siliconangle.com/2026/09/17/ai-data-center-builder-crusoe-valued-at-30-9b-in-3-9b-round/","title":"AI data center builder Crusoe valued at $30.9B in $3.9B round","publisher":"SiliconANGLE (quoting Crusoe)","publishedDate":"2026-09-17","tier":"T5","accessedDate":"2026-10-22","locator":"Crusoe says the 'annualized recurring revenue generated by the service offering topped $100 million within a year of its launch' (Managed Inference)"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$3.56bn a year (the locked card's derivation: Fireworks 1.0, Together 0.45, Baseten 0.57, Modal 0.39, Cerebras 0.25, all others 0.9; range 2.39-5.46). Revenue is counted only for managed serving of other developers' models, net of GPU rental, training and capacity a model developer buys for its own users. The company says Managed Inference ARR topped US$100m (T2; one outlet describes it as contracted ARR). The Thinking Machines agreement of about US$65m a year is a dedicated deployment for a developer's own models (Infrastructure · Cloud) and may sit inside the 100m: counted 30-60-100% = 0.03-0.06-0.10bn."},"media-model-serving":{"low":0,"central":0.111,"high":2.176,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://siliconangle.com/2026/09/17/ai-data-center-builder-crusoe-valued-at-30-9b-in-3-9b-round/","title":"AI data center builder Crusoe valued at $30.9B in $3.9B round","publisher":"SiliconANGLE (quoting Crusoe)","publishedDate":"2026-09-17","tier":"T5","accessedDate":"2026-10-22","locator":"Article (no media split)"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$0.45bn a year (the locked card's derivation: fal 0.35 net, Replicate, Runware and others 0.10; range 0.23-0.97). No media revenue split is disclosed; this is a small allowance (the company may serve none) inside the card's 0.10bn for Replicate, Runware and other hosts. "},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"screening":{"independent-llm-serving":{"levelLow":1,"levelHigh":2},"media-model-serving":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"a7540986e9fa80cef1d3e38ae73735415ea02bea07f3fcc191cdc4b44557f812"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"a7540986e9fa80cef1d3e38ae73735415ea02bea07f3fcc191cdc4b44557f812"}}},"cloudflare-workers-ai":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"cloudflare-workers-ai","name":"Workers AI","kind":"unit","parentId":"cloudflare","commercialActivityVerified":false,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://sdtimes.com/ai/cloudflare-announces-acquisition-of-ai-platform-replicate/","title":"Cloudflare announces acquisition of AI platform Replicate","publisher":"SD Times","publishedDate":"2025-11-17","tier":"T3","accessedDate":"2026-10-22","locator":"Counted in Independent inference providers, as in the card's weight derivation (its allowance names Nebius Token Factory, Crusoe and Cloudflare Workers AI there); the card scope text could also place a neocloud in Cloud model platforms, but each part's weight is proportional to its dollar total, so Reach is almost the same either way"}},"independent-llm-serving":{"low":0.283,"central":1.124,"high":3.297,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://sdtimes.com/ai/cloudflare-announces-acquisition-of-ai-platform-replicate/","title":"Cloudflare announces acquisition of AI platform Replicate","publisher":"SD Times","publishedDate":"2025-11-17","tier":"T5","accessedDate":"2026-10-22","locator":"Cloudflare to acquire Replicate, whose 'over 50,000 production-ready AI models' will be available in Workers AI (17 November 2025); no revenue"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$3.56bn a year (the locked card's derivation: Fireworks 1.0, Together 0.45, Baseten 0.57, Modal 0.39, Cerebras 0.25, all others 0.9; range 2.39-5.46). Revenue is counted only for managed serving of other developers' models, net of GPU rental, training and capacity a model developer buys for its own users. Workers AI language-model serving. Cloudflare reports no Workers AI or Replicate revenue in what was opened. Estimate 0.01-0.04-0.12bn; Replicate's hosting API is counted in Media model hosts below."},"media-model-serving":{"low":2.273,"central":4.444,"high":15.686,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://sdtimes.com/ai/cloudflare-announces-acquisition-of-ai-platform-replicate/","title":"Cloudflare announces acquisition of AI platform Replicate","publisher":"SD Times","publishedDate":"2025-11-17","tier":"T5","accessedDate":"2026-10-22","locator":"Replicate: 50,000+ production-ready models (many image, video and audio)"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$0.45bn a year (the locked card's derivation: fal 0.35 net, Replicate, Runware and others 0.10; range 0.23-0.97). Replicate (consolidated into this entry) hosts mainly image, video and audio models; no revenue is published. Estimate 0.01-0.02-0.08bn, inside the card's 0.05-0.10-0.25bn allowance for Replicate, Runware and others. "},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"screening":{"independent-llm-serving":{"levelLow":1,"levelHigh":3},"media-model-serving":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"a691314e512ef714317650eb1c274988fc51e9c59eed3ab3fac19e1a6c041ebe"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"a691314e512ef714317650eb1c274988fc51e9c59eed3ab3fac19e1a6c041ebe"}}},"hugging-face":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"hugging-face","name":"Hugging Face","kind":"private","commercialActivityVerified":true,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://endpoints.huggingface.co/","title":"Inference Endpoints","publisher":"Hugging Face","publishedDate":"undated","tier":"T2","accessedDate":"2026-10-22","locator":"Inference specialist, not a hyperscaler, enterprise-software or data platform: its managed serving is counted in Independent inference providers"}},"independent-llm-serving":{"low":0.284,"central":1.404,"high":3.306,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"August 2026, annualised","periodEnd":"2026-08-31","source":{"url":"https://techcrunch.com/2026/09/03/nvidia-confirms-it-will-buy-hugging-face-for-12-9-billion/","title":"Nvidia confirms it will buy Hugging Face for $12.9 billion","publisher":"TechCrunch","publishedDate":"2026-09-03","tier":"T5","accessedDate":"2026-10-22","locator":"'Last month, The Information reported that Hugging Face is clocking $150 million in annualized revenue' (3 September 2026); Nvidia confirmed an agreement to buy it for US$12.93bn"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$3.56bn a year (the locked card's derivation: Fireworks 1.0, Together 0.45, Baseten 0.57, Modal 0.39, Cerebras 0.25, all others 0.9; range 2.39-5.46). Revenue is counted only for managed serving of other developers' models, net of GPU rental, training and capacity a model developer buys for its own users. Inference Endpoints only (managed, billed per hour; https://endpoints.huggingface.co/); Inference Providers routing is outside scope. Whole-company annualised revenue is about US$150m (The Information via TechCrunch, T3), mostly subscriptions, enterprise licences and compute, with no endpoint split. Endpoints are assumed to be 7-33-80% of that (no split is published) = 0.01-0.05-0.12bn. The Nvidia agreement had not closed in the sources seen, so Hugging Face stays separate."},"media-model-serving":{"low":0,"central":0.444,"high":6.276,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://endpoints.huggingface.co/","title":"Inference Endpoints","publisher":"Hugging Face","publishedDate":"undated","tier":"T5","accessedDate":"2026-10-22","locator":"Endpoints price list includes an image model (flux-3-image-klein)"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$0.45bn a year (the locked card's derivation: fal 0.35 net, Replicate, Runware and others 0.10; range 0.23-0.97). No media revenue split is disclosed; this is a small allowance (the company may serve none) inside the card's 0.10bn for Replicate, Runware and other hosts. "},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"baselineRevenueUsd":{"low":150000000,"central":150000000,"high":150000000,"unit":"USD","basis":"Annualised company revenue, all products, US dollars","period":"August 2026, annualised","periodEnd":"2026-08-31","source":{"url":"https://techcrunch.com/2026/09/03/nvidia-confirms-it-will-buy-hugging-face-for-12-9-billion/","title":"Nvidia confirms it will buy Hugging Face for $12.9 billion","publisher":"TechCrunch","publishedDate":"2026-09-03","tier":"T3","accessedDate":"2026-10-22","locator":"Whole-company annualised revenue of about US$150m (The Information, as reported by TechCrunch on 3 Sep 2026)"},"note":"About US$150m, as reported; mostly subscriptions, enterprise licences and compute, with no endpoint split. The Nvidia agreement had not closed in the sources seen."},"screening":{"independent-llm-serving":{"levelLow":1,"levelHigh":3},"media-model-serving":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"435e1486602be89874863d71ecc64255eb648ca5a9bb922568e0bbc711a56fdc"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"435e1486602be89874863d71ecc64255eb648ca5a9bb922568e0bbc711a56fdc"}}},"nebius":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"nebius","name":"Nebius (Token Factory)","kind":"public","commercialActivityVerified":false,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://nebius.com/newsroom/nebius-reports-second-quarter-2026-financial-results","title":"Nebius reports second quarter 2026 financial results","publisher":"Nebius","publishedDate":"2026-08-12","tier":"T2","accessedDate":"2026-10-22","locator":"Counted in Independent inference providers, as in the card's weight derivation (its allowance names Nebius Token Factory, Crusoe and Cloudflare Workers AI there); the card scope text could also place a neocloud in Cloud model platforms, but each part's weight is proportional to its dollar total, so Reach is almost the same either way"}},"independent-llm-serving":{"low":0.284,"central":1.404,"high":6.649,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://nebius.com/newsroom/nebius-reports-second-quarter-2026-financial-results","title":"Nebius reports second quarter 2026 financial results","publisher":"Nebius","publishedDate":"2026-08-12","tier":"T5","accessedDate":"2026-10-22","locator":"Token Factory listed as a product; the page links to the results PDF and the shareholder letter (the PDF could not be read as text here)"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$3.56bn a year (the locked card's derivation: Fireworks 1.0, Together 0.45, Baseten 0.57, Modal 0.39, Cerebras 0.25, all others 0.9; range 2.39-5.46). Revenue is counted only for managed serving of other developers' models, net of GPU rental, training and capacity a model developer buys for its own users. Token Factory (managed inference of open models, never GPU rental). Nebius reports AI-cloud run-rate as a whole (US$3.0bn at end of June 2026, seen in search summaries; not opened) and gives no Token Factory revenue; search summaries of the Q2 letter say inference workloads on Token Factory more than tripled in Q2. Token Factory taken as 0.3-1.7-8% of the AI-cloud run-rate = 0.01-0.05-0.25bn. commercialActivityVerified is false because no opened source shows Token Factory revenue."},"media-model-serving":{"low":0,"central":0.111,"high":2.176,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://nebius.com/newsroom/nebius-reports-second-quarter-2026-financial-results","title":"Nebius reports second quarter 2026 financial results","publisher":"Nebius","publishedDate":"2026-08-12","tier":"T5","accessedDate":"2026-10-22","locator":"Results page (no media split)"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$0.45bn a year (the locked card's derivation: fal 0.35 net, Replicate, Runware and others 0.10; range 0.23-0.97). No media revenue split is disclosed; this is a small allowance (the company may serve none) inside the card's 0.10bn for Replicate, Runware and other hosts. "},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"baselineRevenueUsd":{"low":2329200000,"central":2329200000,"high":2329200000,"unit":"USD","basis":"Total revenue, latest quarter × 4, US dollars","period":"Q2 2026 (quarter to 30 June 2026), annualised","periodEnd":"2026-06-30","source":{"url":"https://www.storagenewsletter.com/2026/08/20/nebius-fiscal-2q26-financial-results/","title":"Nebius: Fiscal 2Q26 Financial Results","publisher":"StorageNewsletter (Nebius results release)","publishedDate":"2026-08-20","tier":"T2","accessedDate":"2026-10-22","locator":"Revenue US$582.3 million in Q2 2026 against US$105.1 million in Q2 2025; the page gives no Token Factory or AI-cloud-only figure"},"note":"US$582.3m × 4, the whole company (AI cloud plus Avride and TripleTen); Token Factory revenue is not reported separately and none was found."},"screening":{"independent-llm-serving":{"levelLow":1,"levelHigh":3},"media-model-serving":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"c609bc1dc3f7340fc533f5e61cab1e3b8ba898c6348d166e545be2ff4bb89a78"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"c609bc1dc3f7340fc533f5e61cab1e3b8ba898c6348d166e545be2ff4bb89a78"}}},"ppio":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"ppio","name":"PPIO","kind":"private","commercialActivityVerified":true,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://www.36kr.com/p/3848636043590528","title":"派欧云 (PPIO) 招股书数据 (PPIO prospectus data)","publisher":"36Kr (reporting PPIO's Hong Kong listing prospectus)","publishedDate":"2026-06-11","tier":"T3","accessedDate":"2026-10-22","locator":"China-based edge and AI cloud; its model serving is counted in Model serving in mainland China"}},"independent-llm-serving":{"participates":false,"evidence":{"url":"https://www.36kr.com/p/3848636043590528","title":"派欧云 (PPIO) 招股书数据 (PPIO prospectus data)","publisher":"36Kr (reporting PPIO's Hong Kong listing prospectus)","publishedDate":"2026-06-11","tier":"T3","accessedDate":"2026-10-22","locator":"China-based; serving counted in Model serving in mainland China (card scope)"}},"media-model-serving":{"participates":false,"evidence":{"url":"https://www.36kr.com/p/3848636043590528","title":"派欧云 (PPIO) 招股书数据 (PPIO prospectus data)","publisher":"36Kr (reporting PPIO's Hong Kong listing prospectus)","publishedDate":"2026-06-11","tier":"T3","accessedDate":"2026-10-22","locator":"China-based; serving counted in Model serving in mainland China (card scope)"}},"china-serving":{"low":2,"central":5,"high":12,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"CY2025 revenue, share carried to 2026","periodEnd":"2025-12-31","source":{"url":"https://www.36kr.com/p/3848636043590528","title":"派欧云 (PPIO) 招股书数据 (PPIO prospectus data)","publisher":"36Kr (reporting PPIO's Hong Kong listing prospectus)","publishedDate":"2026-06-11","tier":"T5","accessedDate":"2026-10-22","locator":"2025 total revenue RMB 770m, of which AI cloud computing services RMB 119m (15.5%), gross margin -10.7%; about 1.03 trillion tokens a day as of April 2026"},"note":"5LayerCake estimate (T5). Part total: US$0.92bn a year (the locked card's derivation: IDC's Q3 2026 China MaaS run-rate of about RMB 22bn × a 30% third-party share ÷ 7.15; range 0.52-1.76). Company share = its IDC share of China public-cloud MaaS revenue × the fraction of that revenue from other developers' models ÷ 0.30. Prospectus figures reported by 36Kr (T3; the filing itself not opened). AI cloud revenue RMB 119m ÷ IDC's RMB 3.07bn market = 3.9% × about 45% that is model serving of other developers' open models (the rest is GPU rental) ÷ 0.30 = 5.8%; shown as 5% (2-12%). Any link to Novita is unconfirmed."}},"screening":{"china-serving":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"4d9b856a129a31c15d9d4d0becfc367c65880d5f82f9dd7d04c787de8f1adafd"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"4d9b856a129a31c15d9d4d0becfc367c65880d5f82f9dd7d04c787de8f1adafd"}}},"ibm":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"ibm","name":"IBM (watsonx.ai)","kind":"public","commercialActivityVerified":false,"shares":{"cloud-model-platforms":{"low":0.093,"central":0.37,"high":1.101,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.ibm.com/products/watsonx-ai/foundation-models","title":"Foundation models in watsonx.ai","publisher":"IBM","publishedDate":"undated","tier":"T5","accessedDate":"2026-10-22","locator":"watsonx.ai serves Meta Llama, Mistral, OpenAI gpt-oss, DeepSeek-R1 distills and others alongside IBM Granite, pay-as-you-go or deploy on demand"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$10.82bn a year (the locked card's derivation: Claude sold through cloud platforms, net of the developer's share, 3.76-4.52-5.38bn; OpenAI models sold through Azure, net, 2.4-4.8-9.6bn; other third-party and open models on platforms 0.8-1.5-3.0bn). watsonx.ai. IBM discloses no serving revenue; its 'generative AI book of business' is mostly consulting (IBM's 2026 proxy statement, seen in search results but not opened). Allowance within the card's 'other models on platforms': 0.01-0.04-0.12bn."},"independent-llm-serving":{"participates":false,"evidence":{"url":"https://www.ibm.com/products/watsonx-ai/foundation-models","title":"Foundation models in watsonx.ai","publisher":"IBM","publishedDate":"undated","tier":"T2","accessedDate":"2026-10-22","locator":"Software and cloud platform: counted in Cloud model platforms"}},"media-model-serving":{"participates":false,"evidence":{"url":"https://www.ibm.com/products/watsonx-ai/foundation-models","title":"Foundation models in watsonx.ai","publisher":"IBM","publishedDate":"undated","tier":"T2","accessedDate":"2026-10-22","locator":"Cloud or data platform: image, video and audio models it serves are counted in Cloud model platforms under the card scope, not in this part"}},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"screening":{"cloud-model-platforms":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"94a52b31d1b29a0ae6bef4ca0f2b8afe5d2fe1500ebc812abd44c9c1d3b62794"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"94a52b31d1b29a0ae6bef4ca0f2b8afe5d2fe1500ebc812abd44c9c1d3b62794"}}},"china-telecom-cloud":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"china-telecom-cloud","name":"China Telecom Cloud (eSurfing)","kind":"unit","parentId":"china-telecom","commercialActivityVerified":true,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"Public cloud in China: its serving of other developers' models is counted in Model serving in mainland China, as the card scope states that all serving in mainland China counts there"}},"independent-llm-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"Public cloud in China: its serving of other developers' models is counted in Model serving in mainland China, as the card scope states that all serving in mainland China counts there"}},"media-model-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"Public cloud in China: its serving of other developers' models is counted in Model serving in mainland China, as the card scope states that all serving in mainland China counts there"}},"china-serving":{"low":1,"central":4,"high":12,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T5","accessedDate":"2026-10-22","locator":"China public-cloud MaaS 2025H2 tracker (2025 data): named among public-cloud vendors, no rank given; MaaS revenue RMB 3.07bn in 2025, about RMB 18.6bn forecast for 2026"},"note":"5LayerCake estimate (T5). Part total: US$0.92bn a year (the locked card's derivation: IDC's Q3 2026 China MaaS run-rate of about RMB 22bn × a 30% third-party share ÷ 7.15; range 0.52-1.76). Company share = its IDC share of China public-cloud MaaS revenue × the fraction of that revenue from other developers' models ÷ 0.30. Third-party models only. Share = 2% (1-2-4%: named, outside the top five by revenue, size assumed) × 60% (30-60-90%: fraction from other developers' models) ÷ 0.30 = 4%."}},"screening":{"china-serving":{"levelLow":2,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"7869822739d6c482b9bb803255bdf2f819b78d1287f974cbacf64e7a6a58cc14"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"7869822739d6c482b9bb803255bdf2f819b78d1287f974cbacf64e7a6a58cc14"}}},"runpod":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"runpod","name":"Runpod","kind":"private","commercialActivityVerified":false,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://www.runpod.io/press/runpod-ai-cloud-surpasses-120m-in-arr","title":"Runpod AI cloud surpasses $120M in ARR","publisher":"Runpod","publishedDate":"2026-01-20","tier":"T2","accessedDate":"2026-10-22","locator":"Inference specialist, not a hyperscaler, enterprise-software or data platform: its managed serving is counted in Independent inference providers"}},"independent-llm-serving":{"low":0.141,"central":0.843,"high":2.755,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"ARR stated 20 January 2026, serving part estimated","periodEnd":"2026-01-20","source":{"url":"https://www.runpod.io/press/runpod-ai-cloud-surpasses-120m-in-arr","title":"Runpod AI cloud surpasses $120M in ARR","publisher":"Runpod","publishedDate":"2026-01-20","tier":"T5","accessedDate":"2026-10-22","locator":"'surpassed $120 million in annual recurring revenue'; core offerings GPU Instances, Serverless GPUs, Clusters; no split"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$3.56bn a year (the locked card's derivation: Fireworks 1.0, Together 0.45, Baseten 0.57, Modal 0.39, Cerebras 0.25, all others 0.9; range 2.39-5.46). Revenue is counted only for managed serving of other developers' models, net of GPU rental, training and capacity a model developer buys for its own users. The company says ARR passed US$120m in January 2026 (a later release of 13 October 2026, https://www.runpod.io/press/runpod-325m-arr-new-cfo, opened 22 October 2026, says ARR passed US$325m; like the January release it gives no split by product and says nothing about revenue from running other developers' models). Almost all of it is GPU rental or customers' own code on Serverless GPUs, which counts in Infrastructure · Cloud; only public model endpoints count here. Estimate 0.005-0.03-0.10bn. commercialActivityVerified is false: no opened source shows revenue from running others' models. Both Runpod releases (20 January and 13 October 2026) give only total ARR with no product split; the January release names GPU Instances, Serverless GPUs and Clusters as the core offerings. So this 0.005-0.03-0.10bn range is a T5 allowance with no sourced input."},"media-model-serving":{"low":0,"central":0.889,"high":10.081,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.runpod.io/press/runpod-ai-cloud-surpasses-120m-in-arr","title":"Runpod AI cloud surpasses $120M in ARR","publisher":"Runpod","publishedDate":"2026-01-20","tier":"T5","accessedDate":"2026-10-22","locator":"Press release (no media split)"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$0.45bn a year (the locked card's derivation: fal 0.35 net, Replicate, Runware and others 0.10; range 0.23-0.97). No media revenue split is disclosed; this is a small allowance (the company may serve none) inside the card's 0.10bn for Replicate, Runware and other hosts. "},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"baselineRevenueUsd":{"low":120000000,"central":120000000,"high":120000000,"unit":"USD","basis":"Annualised recurring revenue, all products, US dollars","period":"ARR stated 20 January 2026","periodEnd":"2026-01-20","source":{"url":"https://www.runpod.io/press/runpod-ai-cloud-surpasses-120m-in-arr","title":"Runpod AI cloud surpasses $120M in ARR","publisher":"Runpod","publishedDate":"2026-01-20","tier":"T2","accessedDate":"2026-10-22","locator":"Company press release: ARR above US$120m (20 January 2026)"},"note":"The company says ARR passed US$120m, almost all GPU rental and customers' own code on Serverless GPUs. Its 13 October 2026 release (https://www.runpod.io/press/runpod-325m-arr-new-cfo, opened 22 October 2026) says ARR passed US$325m, again with no split by product; neither figure isolates revenue from serving other developers' models, so this all-products baseline is kept at the January figure and is low against the October run-rate."},"screening":{"independent-llm-serving":{"levelLow":1,"levelHigh":2},"media-model-serving":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"6b16009261c260a8c685acf94e955db07270e701768caa1168770fc0741b81a3"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"6b16009261c260a8c685acf94e955db07270e701768caa1168770fc0741b81a3"}}},"sambanova":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"sambanova","name":"SambaNova (SambaCloud)","kind":"private","commercialActivityVerified":false,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://techcrunch.com/2026/07/08/sambanova-draws-1b-at-11b-valuation-in-series-f-first-close","title":"SambaNova draws $1B at $11B valuation in Series F first close","publisher":"TechCrunch","publishedDate":"2026-07-08","tier":"T3","accessedDate":"2026-10-22","locator":"Inference specialist, not a hyperscaler, enterprise-software or data platform: its managed serving is counted in Independent inference providers"}},"independent-llm-serving":{"low":0.141,"central":0.843,"high":3.288,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://techcrunch.com/2026/07/08/sambanova-draws-1b-at-11b-valuation-in-series-f-first-close","title":"SambaNova draws $1B at $11B valuation in Series F first close","publisher":"TechCrunch","publishedDate":"2026-07-08","tier":"T5","accessedDate":"2026-10-22","locator":"$1bn Series F first close at an $11bn valuation (8 July 2026); JPMorganChase named as 'inference-infrastructure partner' using SN40L and SN50 systems; no revenue or SambaCloud figure"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$3.56bn a year (the locked card's derivation: Fireworks 1.0, Together 0.45, Baseten 0.57, Modal 0.39, Cerebras 0.25, all others 0.9; range 2.39-5.46). Revenue is counted only for managed serving of other developers' models, net of GPU rental, training and capacity a model developer buys for its own users. SambaCloud per-token serving. The article discloses no SambaCloud revenue or customers; most of the company's business is systems sold to customers (JPMorganChase's is on-premises). Estimate 0.005-0.03-0.12bn. Custom inference chips allow a higher replacement level."},"media-model-serving":{"low":0,"central":0.111,"high":2.176,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://techcrunch.com/2026/07/08/sambanova-draws-1b-at-11b-valuation-in-series-f-first-close","title":"SambaNova draws $1B at $11B valuation in Series F first close","publisher":"TechCrunch","publishedDate":"2026-07-08","tier":"T5","accessedDate":"2026-10-22","locator":"Article (no media split)"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$0.45bn a year (the locked card's derivation: fal 0.35 net, Replicate, Runware and others 0.10; range 0.23-0.97). No media revenue split is disclosed; this is a small allowance (the company may serve none) inside the card's 0.10bn for Replicate, Runware and other hosts. "},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"screening":{"independent-llm-serving":{"levelLow":2,"levelHigh":4},"media-model-serving":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"62fc49d27a30db4f0c2618af6dc08d8dc0075c5a7f4c53821c6642b141c20438"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"62fc49d27a30db4f0c2618af6dc08d8dc0075c5a7f4c53821c6642b141c20438"}}},"sap":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"sap","name":"SAP (generative AI hub)","kind":"public","commercialActivityVerified":false,"shares":{"cloud-model-platforms":{"low":0.046,"central":0.277,"high":0.918,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://news.sap.com/netherlands/2026/07/sap-zet-sterke-groei-in-cloud-en-omzet-voort-in-tweede-kwartaal-van-2026/","title":"SAP zet sterke groei in cloud en omzet voort in tweede kwartaal van 2026","publisher":"SAP","publishedDate":"2026-07-24","tier":"T5","accessedDate":"2026-10-22","locator":"Klein: 'strong growth in both our Autonomous Suite and our Business AI Platform' (24 July 2026); the article names no model providers or AI revenue"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$10.82bn a year (the locked card's derivation: Claude sold through cloud platforms, net of the developer's share, 3.76-4.52-5.38bn; OpenAI models sold through Azure, net, 2.4-4.8-9.6bn; other third-party and open models on platforms 0.8-1.5-3.0bn). Generative AI hub in SAP AI Core, billed in capacity units. sap.com and community.sap.com returned 403 (stopped on those hosts) and help.sap.com returned no text, so the model list and billing were seen only in search snippets (OpenAI, Anthropic and Mistral models; consumption measured in GenAI tokens); not opened. Allowance within the card's 'other models on platforms': 0.005-0.03-0.10bn."},"independent-llm-serving":{"participates":false,"evidence":{"url":"https://news.sap.com/netherlands/2026/07/sap-zet-sterke-groei-in-cloud-en-omzet-voort-in-tweede-kwartaal-van-2026/","title":"SAP zet sterke groei in cloud en omzet voort in tweede kwartaal van 2026","publisher":"SAP","publishedDate":"2026-07-24","tier":"T2","accessedDate":"2026-10-22","locator":"Enterprise-software platform: counted in Cloud model platforms"}},"media-model-serving":{"participates":false,"evidence":{"url":"https://news.sap.com/netherlands/2026/07/sap-zet-sterke-groei-in-cloud-en-omzet-voort-in-tweede-kwartaal-van-2026/","title":"SAP zet sterke groei in cloud en omzet voort in tweede kwartaal van 2026","publisher":"SAP","publishedDate":"2026-07-24","tier":"T2","accessedDate":"2026-10-22","locator":"Cloud or data platform: image, video and audio models it serves are counted in Cloud model platforms under the card scope, not in this part"}},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"screening":{"cloud-model-platforms":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"91bf56ebc531c10bb5a93d26e7a287dc9838b657fc8930cf9d16c6e253ae0659"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"91bf56ebc531c10bb5a93d26e7a287dc9838b657fc8930cf9d16c6e253ae0659"}}},"siliconflow":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"siliconflow","name":"SiliconFlow","kind":"private","commercialActivityVerified":true,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://www.36kr.com/p/3878267177991045","title":"硅基流动招股书数据 (SiliconFlow prospectus data)","publisher":"36Kr (reporting SiliconFlow's Hong Kong listing prospectus)","publishedDate":"2026-07-02","tier":"T3","accessedDate":"2026-10-22","locator":"Independent MaaS provider in China, not a cloud or data platform"}},"independent-llm-serving":{"participates":false,"evidence":{"url":"https://www.36kr.com/p/3878267177991045","title":"硅基流动招股书数据 (SiliconFlow prospectus data)","publisher":"36Kr (reporting SiliconFlow's Hong Kong listing prospectus)","publishedDate":"2026-07-02","tier":"T3","accessedDate":"2026-10-22","locator":"Serves open models mainly in mainland China, so counted in Model serving in mainland China (card scope: all serving in mainland China counts there)"}},"media-model-serving":{"participates":false,"evidence":{"url":"https://www.36kr.com/p/3878267177991045","title":"硅基流动招股书数据 (SiliconFlow prospectus data)","publisher":"36Kr (reporting SiliconFlow's Hong Kong listing prospectus)","publishedDate":"2026-07-02","tier":"T3","accessedDate":"2026-10-22","locator":"Serving counted in Model serving in mainland China"}},"china-serving":{"low":1,"central":3,"high":10,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"CY2025 revenue, share carried to 2026","periodEnd":"2025-12-31","source":{"url":"https://www.36kr.com/p/3878267177991045","title":"硅基流动招股书数据 (SiliconFlow prospectus data)","publisher":"36Kr (reporting SiliconFlow's Hong Kong listing prospectus)","publishedDate":"2026-07-02","tier":"T5","accessedDate":"2026-10-22","locator":"2025 revenue RMB 55.33m; public cloud (MaaS) revenue RMB 29.26m (52.9%), gross margin -119%; 1.5% of the whole China MaaS market, fourth; 578.5bn tokens a day in April 2026"},"note":"5LayerCake estimate (T5). Part total: US$0.92bn a year (the locked card's derivation: IDC's Q3 2026 China MaaS run-rate of about RMB 22bn × a 30% third-party share ÷ 7.15; range 0.52-1.76). Company share = its IDC share of China public-cloud MaaS revenue × the fraction of that revenue from other developers' models ÷ 0.30. Prospectus figures reported by 36Kr (T3; the filing itself was not opened); the ranking list 'first, second, third' in the article is Volcano Engine, Alibaba Cloud and Baidu. RMB 29.26m ÷ IDC's RMB 3.07bn market = 0.95% of MaaS revenue × about 95% from other developers' open models ÷ 0.30 = 3%. 2026 growth may differ from the market's (range 1-10%). Another report (IT Home, 5 July 2026) gives the same total revenue."}},"screening":{"china-serving":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"e42ea2ac9a011347d4be83a2b3ccb0696d443b2a9ad57b8e82a6ef3864c26584"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"e42ea2ac9a011347d4be83a2b3ccb0696d443b2a9ad57b8e82a6ef3864c26584"}}},"novita-ai":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"novita-ai","name":"Novita AI","kind":"private","commercialActivityVerified":false,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://novita.ai/","title":"Novita AI","publisher":"Novita AI","publishedDate":"undated","tier":"T2","accessedDate":"2026-10-22","locator":"Inference specialist, not a hyperscaler, enterprise-software or data platform: its managed serving is counted in Independent inference providers"}},"independent-llm-serving":{"low":0.141,"central":0.562,"high":2.21,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://novita.ai/","title":"Novita AI","publisher":"Novita AI","publishedDate":"undated","tier":"T5","accessedDate":"2026-10-22","locator":"'Run 200+ models through a single API'; per-million-token prices; Dedicated Endpoints; GPU Cloud; no revenue or customer counts"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$3.56bn a year (the locked card's derivation: Fireworks 1.0, Together 0.45, Baseten 0.57, Modal 0.39, Cerebras 0.25, all others 0.9; range 2.39-5.46). Revenue is counted only for managed serving of other developers' models, net of GPU rental, training and capacity a model developer buys for its own users. No revenue is published. Estimate 0.005-0.02-0.08bn. Whether Novita is linked to PPIO is unconfirmed (not found in the sources opened); if the two are one group they should be consolidated."},"media-model-serving":{"low":0,"central":0.444,"high":6.276,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://novita.ai/","title":"Novita AI","publisher":"Novita AI","publishedDate":"undated","tier":"T5","accessedDate":"2026-10-22","locator":"Site (model API; no media split)"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$0.45bn a year (the locked card's derivation: fal 0.35 net, Replicate, Runware and others 0.10; range 0.23-0.97). No media revenue split is disclosed; this is a small allowance (the company may serve none) inside the card's 0.10bn for Replicate, Runware and other hosts. "},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"screening":{"independent-llm-serving":{"levelLow":1,"levelHigh":2},"media-model-serving":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"071886e236b95f4973a6059efc7c7d9548412cf2eca8aa8196025039a864ae4a"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"071886e236b95f4973a6059efc7c7d9548412cf2eca8aa8196025039a864ae4a"}}},"friendli-ai":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"friendli-ai","name":"FriendliAI","kind":"private","commercialActivityVerified":false,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://friendli.ai/","title":"FriendliAI","publisher":"FriendliAI","publishedDate":"undated","tier":"T2","accessedDate":"2026-10-22","locator":"Inference specialist, not a hyperscaler, enterprise-software or data platform: its managed serving is counted in Independent inference providers"}},"independent-llm-serving":{"low":0.085,"central":0.562,"high":1.667,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://friendli.ai/","title":"FriendliAI","publisher":"FriendliAI","publishedDate":"undated","tier":"T5","accessedDate":"2026-10-22","locator":"Model APIs, Dedicated Endpoints, bring-your-own-GPU; customers named include LG AI Research, SK Telecom, Kilo and Upstage; no revenue"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$3.56bn a year (the locked card's derivation: Fireworks 1.0, Together 0.45, Baseten 0.57, Modal 0.39, Cerebras 0.25, all others 0.9; range 2.39-5.46). Revenue is counted only for managed serving of other developers' models, net of GPU rental, training and capacity a model developer buys for its own users. No revenue is published; the company's chief executive said in 2025 that revenue would be 6-7 times 2024's (a news item on a site that is an aggregator subdomain, so not cited). Estimate 0.003-0.02-0.06bn."},"media-model-serving":{"low":0,"central":0.111,"high":2.176,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://friendli.ai/","title":"FriendliAI","publisher":"FriendliAI","publishedDate":"undated","tier":"T5","accessedDate":"2026-10-22","locator":"Site (no media split)"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$0.45bn a year (the locked card's derivation: fal 0.35 net, Replicate, Runware and others 0.10; range 0.23-0.97). No media revenue split is disclosed; this is a small allowance (the company may serve none) inside the card's 0.10bn for Replicate, Runware and other hosts. "},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"screening":{"independent-llm-serving":{"levelLow":1,"levelHigh":3},"media-model-serving":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"cea8ff8cb4c251ecd94706e3b12c100aab438238e2d484871de799d42e3f7306"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"cea8ff8cb4c251ecd94706e3b12c100aab438238e2d484871de799d42e3f7306"}}},"thinking-machines":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"thinking-machines","name":"Thinking Machines (Tinker)","kind":"private","commercialActivityVerified":false,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://thinkingmachines.ai/tinker/pricing/","title":"Tinker pricing","publisher":"Thinking Machines Lab","publishedDate":"2026-10-19","tier":"T2","accessedDate":"2026-10-22","locator":"Inference specialist, not a hyperscaler, enterprise-software or data platform: its managed serving is counted in Independent inference providers"}},"independent-llm-serving":{"low":0.056,"central":0.562,"high":2.21,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://thinkingmachines.ai/tinker/pricing/","title":"Tinker pricing","publisher":"Thinking Machines Lab","publishedDate":"2026-10-19","tier":"T5","accessedDate":"2026-10-22","locator":"Per-million-token 'sample' prices (as of 19 October 2026) for open models from Qwen, Moonshot (Kimi), Z.ai, OpenAI (gpt-oss), DeepSeek and NVIDIA; serverless inference is beta and only for its own Inkling models"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$3.56bn a year (the locked card's derivation: Fireworks 1.0, Together 0.45, Baseten 0.57, Modal 0.39, Cerebras 0.25, all others 0.9; range 2.39-5.46). Revenue is counted only for managed serving of other developers' models, net of GPU rental, training and capacity a model developer buys for its own users. Only sampling from customers' fine-tunes of other developers' open models counts; its own Inkling models count in Open-weight, and training fees are outside the sublayer. A run-rate above US$100m for the whole company appears in an unaudited profile (search snippet, not opened) and is not used. Thinking Machines also pays Crusoe about US$65m a year to serve its own models, which is capacity bought by a developer and counts in Infrastructure · Cloud. Estimate 0.002-0.02-0.08bn. Participation check (full pass): the pricing page lists per-million-token sampling prices for other developers' open models (Qwen, Kimi, GLM, gpt-oss, DeepSeek, Nemotron), so Tinker does run other developers' models for customers, billed per token, and the company stays in this sublayer. It has no standalone serverless service for them: sampling sits inside the training workflow ('forward-only requests from a training client are charged at the training rate'), and its serverless inference is for its own Inkling models only (Open-weight). No sampling revenue is published; if the checker judges sampling inside training not to be serving for others, the function should be declared non-participating."},"media-model-serving":{"low":0,"central":0.111,"high":2.176,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://thinkingmachines.ai/tinker/pricing/","title":"Tinker pricing","publisher":"Thinking Machines Lab","publishedDate":"2026-10-19","tier":"T5","accessedDate":"2026-10-22","locator":"Pricing page (language models only)"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$0.45bn a year (the locked card's derivation: fal 0.35 net, Replicate, Runware and others 0.10; range 0.23-0.97). No media revenue split is disclosed; this is a small allowance (the company may serve none) inside the card's 0.10bn for Replicate, Runware and other hosts. "},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"screening":{"independent-llm-serving":{"levelLow":1,"levelHigh":2},"media-model-serving":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"1a896a6d13edb45ca9ab7b79c7205ff878d334b51739eec479f9f7a32a557f37"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"1a896a6d13edb45ca9ab7b79c7205ff878d334b51739eec479f9f7a32a557f37"}}},"infinigence-ai":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"infinigence-ai","name":"Infinigence AI","kind":"private","commercialActivityVerified":false,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://aiweekly.co/alerts/shanghai-ai-infra-firm-infinigence-closes-rmb-23b-344m-series-c-with-tencent","title":"Shanghai AI-Infra Firm Infinigence Closes RMB 2.3B ($344M) Series C With Tencent, Baidu and Lenovo","publisher":"AI Weekly","publishedDate":"2026-10-13","tier":"T3","accessedDate":"2026-10-22","locator":"China-based AI-infrastructure software company; its model serving is counted in Model serving in mainland China"}},"independent-llm-serving":{"participates":false,"evidence":{"url":"https://aiweekly.co/alerts/shanghai-ai-infra-firm-infinigence-closes-rmb-23b-344m-series-c-with-tencent","title":"Shanghai AI-Infra Firm Infinigence Closes RMB 2.3B ($344M) Series C With Tencent, Baidu and Lenovo","publisher":"AI Weekly","publishedDate":"2026-10-13","tier":"T3","accessedDate":"2026-10-22","locator":"China-based; serving counted in Model serving in mainland China (card scope)"}},"media-model-serving":{"participates":false,"evidence":{"url":"https://aiweekly.co/alerts/shanghai-ai-infra-firm-infinigence-closes-rmb-23b-344m-series-c-with-tencent","title":"Shanghai AI-Infra Firm Infinigence Closes RMB 2.3B ($344M) Series C With Tencent, Baidu and Lenovo","publisher":"AI Weekly","publishedDate":"2026-10-13","tier":"T3","accessedDate":"2026-10-22","locator":"China-based; serving counted in Model serving in mainland China (card scope)"}},"china-serving":{"low":0.2,"central":1.5,"high":6,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"Q3 2026, annualised (5LayerCake estimate)","periodEnd":"2026-09-30","source":{"url":"https://aiweekly.co/alerts/shanghai-ai-infra-firm-infinigence-closes-rmb-23b-344m-series-c-with-tencent","title":"Shanghai AI-Infra Firm Infinigence Closes RMB 2.3B ($344M) Series C With Tencent, Baidu and Lenovo","publisher":"AI Weekly","publishedDate":"2026-10-13","tier":"T5","accessedDate":"2026-10-22","locator":"RMB 2.3bn Series C closed 12 October 2026; software layer for mixed-chip training and inference; confidential Hong Kong IPO filing; no revenue or token figure"},"note":"5LayerCake estimate (T5). Part total: US$0.92bn a year (the locked card's derivation: IDC's Q3 2026 China MaaS run-rate of about RMB 22bn × a 30% third-party share ÷ 7.15; range 0.52-1.76). Company share = its IDC share of China public-cloud MaaS revenue × the fraction of that revenue from other developers' models ÷ 0.30. The article does not describe a MaaS platform; search snippets say revenue comes from token-based fees, compute rental and private-deployment fees, with named model-lab partners (not opened). No revenue is published. Estimate 1.5% (0.2-6%)."}},"screening":{"china-serving":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"c2cd9433d5494466f0880c8e137ec289809a0f66e2503b7fa837bf93a24d0167"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"c2cd9433d5494466f0880c8e137ec289809a0f66e2503b7fa837bf93a24d0167"}}},"chutes":{"schema":"5lc-index-inputs/1","key":"models-inference-serving","entityId":"chutes","name":"Chutes","kind":"private","commercialActivityVerified":false,"shares":{"cloud-model-platforms":{"participates":false,"evidence":{"url":"https://chutes.ai/","title":"Chutes: serverless compute for AI","publisher":"Chutes Global Corp","publishedDate":"undated","tier":"T2","accessedDate":"2026-10-22","locator":"Inference specialist, not a hyperscaler, enterprise-software or data platform: its managed serving is counted in Independent inference providers"}},"independent-llm-serving":{"low":0.037,"central":0.112,"high":0.23,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"H1 2026 run-rate, annualised (5LayerCake estimate)","periodEnd":"2026-06-30","source":{"url":"https://ownyourmind.ai/tokenomics/chutes-bittensor-revenue-machine/","title":"Chutes: Bittensor's Revenue Machine, Subsidised","publisher":"Own Your Mind","publishedDate":"2026-04-10","tier":"T5","accessedDate":"2026-10-22","locator":"Section 'The OpenRouter contradiction': Chutes' own daily revenue series (withdrawn mid-2026), rebuilt from the Wayback Machine: last 90 published days averaged $15.6K a day, peak $22.5K on 25 March 2026, part of it through February sponsored hosting paid by model makers. Page dated 10 April 2026, updated 6 October 2026."},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$3.56bn a year (the locked card's derivation: Fireworks 1.0, Together 0.45, Baseten 0.57, Modal 0.39, Cerebras 0.25, all others 0.9; range 2.39-5.46). Revenue is counted only for managed serving of other developers' models, net of GPU rental, training and capacity a model developer buys for its own users (sponsored hosting paid by model makers is left out). Chutes publishes no revenue now. Inputs, all opened on 22 October 2026. (1) Chutes' own daily revenue series, withdrawn mid-2026 and rebuilt from the Wayback Machine by Own Your Mind (the source above, a blog reconstruction, T4 at best): the last 90 published days averaged US$15.6k a day, with a peak of US$22.5k on 25 March 2026; part of the figure through February was sponsored hosting paid by model makers. The page does not date the window; it contains 25 March 2026 and the series ended 'mid-2026', so the period end is set to 30 June 2026 at the latest. 15.6k × 365 = US$5.69m a year; less 25% sponsored (the 25% split is TAO.media's, input 2) = US$4.27m. (2) TAO.media, 'The Investor's Guide to Chutes', 24 February 2026 (https://www.tao.media/the-investors-guide-to-chutes-bittensors-inference-layer/), T4 at best, no source named for the figure: US$5.5m annualised revenue, 75% organic and 25% sponsored; organic 5.5 × 0.75 = US$4.1m. (3) Pine Analytics, 'The Bear Case for Bittensor (TAO)', 23 March 2026 (https://pineanalytics.substack.com/p/the-bear-case-for-bittensor-tao), T4: external customer revenue about US$1.3-2.4m a year (US$2.4m self-reported by the Chutes team to DL News in October 2025; US$1.3m a March 2026 estimate; neither audited). Range: low 0.0013bn = Pine's low; central 0.004bn = about US$4.2m net serving revenue (inputs 1 and 2 agree within 4%) × 95% for language models (the other 5% is the media allowance in the next part); high 0.0082bn = the peak day of US$22.5k × 365 = US$8.2m, an upper bound because a peak day is not sustained. Shares: 0.0013 ÷ (3.56 − 0.004 + 0.0013) = 0.037%; 0.004 ÷ 3.56 = 0.112%; 0.0082 ÷ (3.56 − 0.004 + 0.0082) = 0.230%. Not used: an earlier draft of this file cited about US$43m (Q1 2026 annualised) to about US$172m, seen only in search snippets. No page opened for this fix carries those figures; the nearest, about US$52m a year, is Pine's value of the TAO emissions paid to the subnet, which is not customer revenue. The 'approaching US$10m ARR' claim is not used either: the Own Your Mind page traces it to a media post, not the company, and finds Chutes' own series well under it. No Q3 2026 revenue was found, so the H1 2026 run-rate stands as the latest. Chutes' share of TAO emissions fell from 14.39% (Pine, March 2026) to 5.82% (Own Your Mind chain read, 22 September 2026); this does not by itself change customer revenue."},"media-model-serving":{"low":0.003,"central":0.044,"high":0.266,"unit":"percent","basis":"Share of the function's annualised net serving revenue in US dollars: customer payments for running models, less any share paid to the model's developer (latest month × 12 or latest quarter × 4)","period":"H1 2026 run-rate, annualised (5LayerCake estimate)","periodEnd":"2026-06-30","source":{"url":"https://ownyourmind.ai/tokenomics/chutes-bittensor-revenue-machine/","title":"Chutes: Bittensor's Revenue Machine, Subsidised","publisher":"Own Your Mind","publishedDate":"2026-04-10","tier":"T5","accessedDate":"2026-10-22","locator":"Same page: the public model list includes diffusion models, embeddings, speech and 3D generation; no media revenue split is given"},"note":"5LayerCake estimate (T5). Share = company ÷ part total, with the other companies held at their central values. Part total: US$0.45bn a year (the locked card's derivation: fal 0.35 net, Replicate, Runware and others 0.10; range 0.23-0.97). The cited page lists diffusion, speech and 3D models among Chutes' 50+ models but gives no media revenue, and Chutes' reported volume is overwhelmingly language-model tokens. Allowance (T5 judgment) as a share of the Chutes net serving revenue range in the language-model part of this file (inputs and arithmetic there): low 1% × US$1.3m = US$13k (0.000013bn); central 5% × US$4.2m = US$0.21m (0.0002bn); high 15% × US$8.2m = US$1.2m (0.0012bn). This replaces an earlier allowance of 0 to US$30m that was larger than all of the company's evidenced revenue. Shares: 0.000013 ÷ (0.45 − 0.0002 + 0.000013) = 0.003%; 0.0002 ÷ 0.45 = 0.044%; 0.0012 ÷ (0.45 − 0.0002 + 0.0012) = 0.266%."},"china-serving":{"participates":false,"evidence":{"url":"https://www.idc.com/resource-center/blog/%E4%B8%AD%E5%9B%BDmaas%E5%B8%82%E5%9C%BA%E8%BF%9B%E5%85%A5%E9%AB%98%E9%80%9F%E5%A2%9E%E9%95%BF%E6%9C%9F%EF%BC%8Ctoken%E7%BB%8F%E6%B5%8E%E4%BB%8E%E6%A6%82%E5%BF%B5%E8%B5%B0%E5%90%91%E8%A7%84%E6%A8%A1/","title":"中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践 (China's MaaS market enters a high-growth phase)","publisher":"IDC","publishedDate":"2026-05-07","tier":"T4","accessedDate":"2026-10-22","locator":"IDC's lists of China public-cloud MaaS vendors (ranked and named): the company is not among them, and serving outside mainland China is not in this part"}}},"screening":{"independent-llm-serving":{"levelLow":1,"levelHigh":2},"media-model-serving":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"2fd28810d0c5707da13a8c4c9456de0d55ebb9e31d43e11eae4558401617554b"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"2fd28810d0c5707da13a8c4c9456de0d55ebb9e31d43e11eae4558401617554b"}}}},"ratings":{"microsoft-intelligent-cloud":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"microsoft-intelligent-cloud","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"cloud-model-platforms":{"alternatives":{"level":3,"evidence":[{"url":"https://blogs.microsoft.com/blog/2026/04/27/the-next-phase-of-the-microsoft-openai-partnership/","date":"2026-04-27","support":"Microsoft: OpenAI can now serve all its products to customers across any cloud provider; licence now non-exclusive."},{"url":"https://www.aboutamazon.com/news/aws/bedrock-openai-models","date":"2026-09-08","support":"Amazon (page updated 8 Sep 2026): GPT-6 Astra and the GPT-5.6 family are generally available on Bedrock. Google Vertex lists only gpt-oss; few full alternatives (Bedrock, OpenAI direct)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://aws.amazon.com/about-aws/whats-new/2026/06/amazon-bedrock-openai-models-codex-generally-available/","date":"2026-06-01","support":"AWS: GPT-5.5/5.4 generally available on Bedrock; pricing matches OpenAI first-party rates; usage counts toward AWS commitments."},{"url":"https://devblogs.microsoft.com/azuregov/azure-openai-authorization/","date":"2025-04-16","support":"Microsoft: Azure OpenAI is authorised for DoD IL6 and Top Secret clouds; classified volume would take longer to move than commercial volume (3-12 months)."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://press.aboutamazon.com/2026/2/openai-and-amazon-announce-strategic-partnership","date":"2026-02-27","support":"Amazon: OpenAI commits to about 2 gigawatts of Trainium capacity through AWS; the best alternative has large OpenAI-model capacity."},{"url":"https://blogs.microsoft.com/blog/2026/04/27/the-next-phase-of-the-microsoft-openai-partnership/","date":"2026-04-27","support":"Microsoft: OpenAI products ship first on Azure unless Microsoft cannot support them; alternatives may get new models later (small gap)."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"microsoft-intelligent-cloud","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"cloud-model-platforms":{"alternatives":{"level":2,"evidence":[{"url":"https://blogs.microsoft.com/blog/2026/04/27/the-next-phase-of-the-microsoft-openai-partnership/","date":"2026-04-27","support":"Microsoft says OpenAI can now serve all its products on any cloud provider (27 Apr 2026), ending Azure's exclusive route to OpenAI models."},{"url":"https://aws.amazon.com/blogs/machine-learning/openai-models-and-codex-on-amazon-bedrock-are-now-generally-available/","date":"2026-06-01","support":"AWS says GPT-5.5, GPT-5.4 and Codex are generally available on Amazon Bedrock (byline date June 1, 2026)."},{"url":"https://investors.snowflake.com/news/news-details/2026/Snowflake-and-OpenAI-Forge-200-Million-Partnership-to-Bring-Enterprise-Ready-AI-to-the-Worlds-Most-Trusted-Data-Platform/default.aspx","date":"2026-02-02","support":"Snowflake announced a multi-year OpenAI partnership putting OpenAI models natively in Cortex AI across all three major clouds (2 Feb 2026)."},{"url":"https://docs.databricks.com/aws/en/machine-learning/model-serving/foundation-model-overview","date":"2026-10-19","support":"Databricks lists GPT-6 and GPT-5.x models as hosted (page updated Oct 19, 2026)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://aws.amazon.com/blogs/machine-learning/openai-models-and-codex-on-amazon-bedrock-are-now-generally-available/","date":"2026-06-01","support":"Bedrock offered OpenAI models generally about five weeks after Microsoft's 27 April 2026 announcement, so a rival platform can be ready within weeks; enterprise approvals and commitments take longer (inference)."},{"url":"https://a16z.com/ai-enterprise-2025/","date":"2025-06-10","support":"a16z's CIO survey says switching costs are rising and changing models can take a lot of engineering time; buyers go direct or via Databricks when their main cloud lacks a model (10 Jun 2025)."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://blogs.microsoft.com/blog/2026/04/27/the-next-phase-of-the-microsoft-openai-partnership/","date":"2026-04-27","support":"Microsoft says it remains OpenAI's primary cloud partner and OpenAI products ship first on Azure unless it cannot or chooses not to support them: a lead over Bedrock."},{"url":"https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/foundry-models-overview","date":"2026-07-28","support":"Microsoft says the Foundry catalog has over 10,000 models, including Anthropic, DeepSeek, Meta, Mistral and Cohere (page metadata date 2026-07-28)."},{"url":"https://docs.aws.amazon.com/bedrock/latest/userguide/model-cards.html","date":"undated","support":"AWS's model page (no date) lists 19 providers, including OpenAI GPT-6 and Anthropic Claude 5.x, a narrower catalog than Foundry's."}]}}}}},"aws":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"aws","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"cloud-model-platforms":{"alternatives":{"level":3,"evidence":[{"url":"https://blogs.microsoft.com/blog/2025/11/18/microsoft-nvidia-and-anthropic-announce-strategic-partnerships/","date":"2025-11-18","support":"Microsoft says Claude becomes the only frontier model on all three leading clouds, via Microsoft Foundry."},{"url":"https://platform.claude.com/docs/en/build-with-claude/claude-models-on-google-cloud/google-managed","date":"undated","support":"Anthropic docs (undated): Google Cloud serves current Claude models; requests are nearly identical to the Messages API."},{"url":"https://blogs.microsoft.com/blog/2026/04/27/the-next-phase-of-the-microsoft-openai-partnership/","date":"2026-04-27","support":"Microsoft: OpenAI products ship first on Azure, a full OpenAI platform; with Google, two hyperscale peers to Bedrock."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://platform.claude.com/docs/en/build-with-claude/claude-models-on-google-cloud/google-managed","date":"undated","support":"Anthropic docs (undated): moving Claude calls to Google Cloud changes client, endpoint and auth; some features unsupported. Code change is small, approvals are not."},{"url":"https://aws.amazon.com/about-aws/whats-new/2026/06/amazon-bedrock-openai-models-codex-generally-available/","date":"2026-06-01","support":"AWS: Bedrock usage counts toward existing AWS commitments, so customers moving volume elsewhere also re-do procurement and security reviews (judgment: 3-12 months)."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.anthropic.com/news/expanding-our-use-of-google-cloud-tpus-and-services","date":"2025-10-23","support":"Anthropic says it is expanding to up to one million Google TPUs, well over a gigawatt online in 2026: large Claude capacity outside AWS."},{"url":"https://www.anthropic.com/news/claude-in-amazon-bedrock-fedramp-high","date":"2025-06-11","support":"Anthropic: Claude on Bedrock GovCloud is approved for FedRAMP High and DoD IL4/5; Vertex listed for FedRAMP High and IL2 only, a small gap for government volume."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"aws","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"cloud-model-platforms":{"alternatives":{"level":2,"evidence":[{"url":"https://azure.microsoft.com/en-us/blog/claude-in-microsoft-foundry-is-now-generally-available/","date":"2026-06-29","support":"Claude, the main third-party line on Bedrock, is generally available on Microsoft Foundry too; Microsoft says Anthropic operates the inference there (page date 2026-06-29)."},{"url":"https://platform.claude.com/docs/en/build-with-claude/claude-models-on-google-cloud/google-managed","date":"undated","support":"Anthropic's documentation (no date) describes Claude offered as a Google-run service on Google Cloud, a second hyperscaler alternative."},{"url":"https://docs.databricks.com/aws/en/machine-learning/model-serving/foundation-model-overview","date":"2026-10-19","support":"Databricks also hosts Claude 5.x and GPT-6 models (page updated Oct 19, 2026), so four or more other platforms serve the same lines."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://a16z.com/ai-enterprise-2025/","date":"2025-06-10","support":"a16z's CIO survey says switching costs are rising and changing models can take a lot of engineering time; buyers go direct or via Databricks when their main cloud lacks a model (10 Jun 2025)."},{"url":"https://blogs.microsoft.com/blog/2025/11/18/microsoft-nvidia-and-anthropic-announce-strategic-partnerships/","date":"2025-11-18","support":"Microsoft announced Claude for Foundry customers on 2025-11-18, with Anthropic committing $30bn of Azure compute."},{"url":"https://azure.microsoft.com/en-us/blog/claude-in-microsoft-foundry-is-now-generally-available/","date":"2026-06-29","support":"Azure's blog dates general availability of Claude on Foundry to 2026-06-29, about seven months after the announcement: a rough guide to how long a platform needs to qualify and ramp a model line."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://platform.claude.com/docs/en/build-with-claude/claude-models-on-google-cloud/google-managed","date":"undated","support":"Anthropic's documentation (no date) says Claude run by Google lacks the Files API, Message Batches API and several server-side tools, so that alternative is not at full parity."},{"url":"https://azure.microsoft.com/en-us/blog/claude-in-microsoft-foundry-is-now-generally-available/","date":"2026-06-29","support":"Foundry's Claude supports the Messages API with prompt caching and extended thinking (page date 2026-06-29): close to parity, but generally available only since June 2026."}]}}}}},"google-cloud":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"google-cloud","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"cloud-model-platforms":{"alternatives":{"level":3,"evidence":[{"url":"https://blogs.microsoft.com/blog/2025/11/18/microsoft-nvidia-and-anthropic-announce-strategic-partnerships/","date":"2025-11-18","support":"Microsoft: Claude available on Microsoft Foundry, making it available on all three leading clouds; Bedrock and Foundry are the hyperscale alternatives."},{"url":"https://www.khaleejtimes.com/business/anthropic-ipo-prospectus-lays-bare-deep-dependence-on-big-tech-partners","date":"2026-09-30","support":"Reuters: Anthropic's filing says it sells Claude through Amazon, Google and Microsoft cloud platforms."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://platform.claude.com/docs/en/build-with-claude/claude-models-on-google-cloud/google-managed","date":"undated","support":"Anthropic docs (undated): Google Cloud Claude requests differ from the Messages API mainly in endpoint, auth and some unsupported features; enterprise re-approval sets the time (judgment: 3-12 months)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.anthropic.com/news/claude-in-amazon-bedrock-fedramp-high","date":"2025-06-11","support":"Anthropic: Claude on Bedrock GovCloud approved for FedRAMP High and DoD IL4/5, beyond Vertex's FedRAMP High and IL2: the best alternative has no gap."},{"url":"https://aws.amazon.com/about-aws/whats-new/2026/06/amazon-bedrock-openai-models-codex-generally-available/","date":"2026-06-01","support":"AWS: Bedrock also carries OpenAI's GPT-5.5/5.4 in production, which Vertex does not."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"google-cloud","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"cloud-model-platforms":{"alternatives":{"level":2,"evidence":[{"url":"https://docs.aws.amazon.com/bedrock/latest/userguide/model-cards.html","date":"undated","support":"AWS's model page (no date) lists Anthropic Claude 5.x, Meta Llama 4, Mistral, DeepSeek, Qwen and OpenAI models, the third-party lines Google serves."},{"url":"https://azure.microsoft.com/en-us/blog/claude-in-microsoft-foundry-is-now-generally-available/","date":"2026-06-29","support":"Claude is generally available on Microsoft Foundry (page date 2026-06-29)."},{"url":"https://docs.databricks.com/aws/en/machine-learning/model-serving/foundation-model-overview","date":"2026-10-19","support":"Databricks hosts Claude, GPT, Llama and Gemma models (page updated Oct 19, 2026)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://a16z.com/ai-enterprise-2025/","date":"2025-06-10","support":"a16z's CIO survey says switching costs are rising and changing models can take a lot of engineering time; buyers go direct or via Databricks when their main cloud lacks a model (10 Jun 2025)."},{"url":"https://blogs.microsoft.com/blog/2025/11/18/microsoft-nvidia-and-anthropic-announce-strategic-partnerships/","date":"2025-11-18","support":"Microsoft announced Claude for Foundry customers on 2025-11-18, with Anthropic committing $30bn of Azure compute."},{"url":"https://azure.microsoft.com/en-us/blog/claude-in-microsoft-foundry-is-now-generally-available/","date":"2026-06-29","support":"Azure's blog dates general availability of Claude on Foundry to 2026-06-29, about seven months after the announcement: a rough guide to how long a platform needs to qualify and ramp a model line."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://platform.claude.com/docs/en/build-with-claude/claude-models-on-google-cloud/google-managed","date":"undated","support":"Anthropic's documentation (no date) says Claude run by Google lacks the Files API, Message Batches API and several server-side tools, so rivals' offers are not a step down."},{"url":"https://docs.aws.amazon.com/bedrock/latest/userguide/model-cards.html","date":"undated","support":"AWS's model page (no date) shows Bedrock carrying Claude 5.x, Llama 4, Mistral, DeepSeek, Qwen and OpenAI GPT-6, a superset of Google's third-party lines."}]}}}}},"fireworks-ai":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"fireworks-ai","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":2,"evidence":[{"url":"https://www.verdict.co.uk/baseten-secures-1-5bn-series-f/","date":"2026-06-23","support":"Verdict: Baseten handles over one billion inference calls a day across 87 clusters and 18 clouds."},{"url":"https://thenextweb.com/news/together-ai-800m-series-c-aramco-ventures","date":"2026-07-01","support":"TNW: Together AI books over $1bn a year and names Cursor and Cognition as customers; Fireworks also names Cursor, so customers qualify several providers."},{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Benchmark page (no publication date; 72-hour medians, read 2026-10-22) lists 16 API providers for gpt-oss-120b, including Together AI, Baseten, Azure, Google Vertex and Amazon."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://docs.fireworks.ai/tools-sdks/openai-compatibility","date":"undated","support":"Fireworks docs (undated): works with the OpenAI client; switching means changing base URL, key and model name."},{"url":"https://fireworks.ai/blog/series-d-announcement","date":"2026-07-15","support":"The company says customers own their specialised models instead of renting them, so weights can move to another host."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://fireworks.ai/blog/series-d-announcement","date":"2026-07-15","support":"The company says it serves more than 40 trillion tokens a day, over 95% from customer-specialised models, and passes $1bn run-rate."},{"url":"https://www.verdict.co.uk/baseten-secures-1-5bn-series-f/","date":"2026-06-23","support":"Verdict: Baseten says revenue grew about 20x year over year; the nearest rivals are smaller but large: a small gap."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"fireworks-ai","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://huggingface.co/docs/inference-providers/index","date":"undated","support":"Hugging Face documentation (no date) lists 15 chat-model inference partners, including Baseten, Cerebras, DeepInfra, Fireworks, Groq, Novita and Together, behind one API."},{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Independent benchmark page (no date shown for the table) lists 16 API providers for gpt-oss-120b; the fastest are Cerebras 1,815, SambaNova 707 and Groq 472 tokens per second."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://fireworks.ai/blog/series-d-announcement","date":"2026-07-15","support":"The company says over 95% of its 40+ trillion daily tokens come from models specialised on customers' data; moving tuned models needs re-deployment and re-testing (inference)."},{"url":"https://www.together.ai/blog/announcing-our-series-c","date":"2026-07-01","support":"Decagon reduced inference costs sixfold after moving to Together AI (the company's account, 1 Jul 2026): moves between providers happen."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"On the same benchmark page (undated table) CoreWeave, DeepInfra and Novita are the cheapest and Cerebras the fastest, so no alternative is a step down from commodity open-model serving."},{"url":"https://www.together.ai/blog/announcing-our-series-c","date":"2026-07-01","support":"Together says it has commitments for over 500 MW of compute and thousands of customers (1 Jul 2026), so a rival operates at comparable scale (company statement)."},{"url":"https://baseten.co/blog/announcing-our-series-f/","date":"2026-06-22","support":"Baseten says inference volume grew 40x in a year (22 Jun 2026), another rival scaling fast (company statement)."}]}}}}},"fal":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"fal","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"media-model-serving":{"alternatives":{"level":3,"evidence":[{"url":"https://techcrunch.com/2025/12/11/runware-raises-50m-series-a-from-dawn-capital-comcast-ventures-to-become-the-api-for-all-ai","date":"2025-12-11","support":"TechCrunch: Runware names fal and Replicate as its competitors; few hosts offer comparable breadth."},{"url":"https://blog.cloudflare.com/replicate-joins-cloudflare/","date":"2025-11-17","support":"Cloudflare: Replicate, now part of Cloudflare, hosts 50,000+ models and keeps its API running."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://fal.ai/","date":"undated","support":"fal homepage (undated): the company says it scales to 100M+ daily inference calls for 2.5M+ developers; rivals must grow capacity to absorb this (evidence thin; judgment 3-12 months)."},{"url":"https://techcrunch.com/2025/12/11/runware-raises-50m-series-a-from-dawn-capital-comcast-ventures-to-become-the-api-for-all-ai","date":"2025-12-11","support":"TechCrunch: Runware reroutes workloads to third-party clouds when it needs more capacity, so ramp depends on rented GPUs."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://techcrunch.com/2025/12/09/fal-nabs-140m-in-fresh-funding-led-by-sequoia-tripling-valuation-to-4-5b/","date":"2025-12-09","support":"TechCrunch citing Bloomberg: fal passed $200m revenue by October 2025, serving Adobe, Shopify, Canva and Quora."},{"url":"https://techcrunch.com/2025/12/11/runware-raises-50m-series-a-from-dawn-capital-comcast-ventures-to-become-the-api-for-all-ai","date":"2025-12-11","support":"TechCrunch: Runware has about 25 staff and $66m raised in total; alternatives are materially smaller."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"fal","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"media-model-serving":{"alternatives":{"level":2,"evidence":[{"url":"https://huggingface.co/docs/inference-providers/index","date":"undated","support":"Hugging Face documentation (no date) lists text-to-image partners fal, HF Inference, Nscale, Replicate, Together and WaveSpeed, and text-to-video partners fal, Novita, Replicate, Together and WaveSpeed."},{"url":"https://www.helpnetsecurity.com/2025/11/17/cloudflare-to-acquire-replicate/","date":"2025-11-17","support":"Cloudflare said it would add Replicate's 50,000+ models to Workers AI (17 Nov 2025)."},{"url":"https://www.cityam.com/runware-raises-50m-to-accelerate-ai-media-generation/","date":"2025-12-12","support":"Runware, a rival media host, says it serves 200,000 developers and counts Wix and Quora as customers (12 Dec 2025)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://tech.eu/2025/12/09/fal-raises-140m-series-d-to-power-the-next-era-of-real-time-generative-media/","date":"2025-12-09","support":"fal says it serves millions of developers and hundreds of enterprise teams across image, video, audio and 3D (9 Dec 2025); a rival must match that throughput, and video capacity ramps take months (inference, thin evidence)."},{"url":"https://huggingface.co/blog/inference-providers","date":"2025-01-28","support":"Hugging Face's launch providers included fal and Replicate, and switching needed only a changed provider name (28 Jan 2025): small workloads move easily, large ones need capacity."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://www.cityam.com/runware-raises-50m-to-accelerate-ai-media-generation/","date":"2025-12-12","support":"Runware, a rival media host, says it serves 200,000 developers (12 Dec 2025)."},{"url":"https://tech.eu/2025/12/09/fal-raises-140m-series-d-to-power-the-next-era-of-real-time-generative-media/","date":"2025-12-09","support":"fal says it serves millions of developers (9 Dec 2025), roughly ten times Runware's developer count (company statements)."},{"url":"https://www.latent.space/p/fal","date":"undated","support":"On a podcast (date not shown) fal's engineers claim 100+ custom kernels and 1.5-10x speed-ups, covering about 70-80% of diffusion-transformer models (company claims, unverified)."}]}}}}},"baseten":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"baseten","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":2,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Benchmark page (no publication date; 72-hour medians, read 2026-10-22) lists 16 API providers for gpt-oss-120b, including Together AI, DeepInfra, Nebius, Crusoe, Groq and Cerebras."},{"url":"https://thenextweb.com/news/together-ai-800m-series-c-aramco-ventures","date":"2026-07-01","support":"TNW: Together AI books over $1bn a year serving open models for Cursor, Cognition and Decagon."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://docs.baseten.co/overview","date":"undated","support":"Baseten docs (undated): Model APIs support the OpenAI Chat Completions API; custom models deploy on dedicated GPUs, so clients and weights are portable."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://fireworks.ai/blog/series-d-announcement","date":"2026-07-15","support":"The company (Fireworks) says it serves over 40 trillion tokens a day, mostly customer-specialised models: an alternative of at least equal scale."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"baseten","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://huggingface.co/docs/inference-providers/index","date":"undated","support":"Hugging Face documentation (no date) lists 15 chat-model inference partners, including Baseten, Cerebras, DeepInfra, Fireworks, Groq, Novita and Together, behind one API."},{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Independent benchmark page (no date shown for the table) lists 16 API providers for gpt-oss-120b; the fastest are Cerebras 1,815, SambaNova 707 and Groq 472 tokens per second."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://baseten.co/blog/announcing-our-series-f/","date":"2026-06-22","support":"The company says it post-trains and serves specialised models for customers such as Cursor, Notion and Harvey (22 Jun 2026); moving tuned or dedicated deployments needs re-optimisation (inference, thin evidence)."},{"url":"https://huggingface.co/blog/inference-providers","date":"2025-01-28","support":"Hugging Face says switching provider in its client needs only a changed provider name, with the rest of the code unchanged (posted 2025-01-28)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"On the same benchmark page (undated table) CoreWeave, DeepInfra and Novita are the cheapest and Cerebras the fastest, so no alternative is a step down from commodity open-model serving."},{"url":"https://fireworks.ai/blog/series-d-announcement","date":"2026-07-15","support":"Fireworks says it passed a $1bn annualised run rate and serves over 40 trillion tokens a day (15 Jul 2026), so rivals match Baseten's scale (company statement)."}]}}}}},"cerebras-inference":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"cerebras-inference","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":3,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Benchmark page (undated; 72-hour medians, read 2026-10-22): on gpt-oss-120b only SambaNova (707 t/s) and Groq (472 t/s) approach Cerebras (1,815 t/s)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://inference-docs.cerebras.ai/resources/openai","date":"undated","support":"Cerebras docs (undated): OpenAI-compatible; existing apps switch by changing API key, base URL and model."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Benchmark page (undated): best alternative SambaNova runs gpt-oss-120b at about 39% of Cerebras's output speed; GPU providers peak near 382 t/s. Material gap for speed-bound users."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"cerebras-inference","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":3,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"On gpt-oss-120b Cerebras runs 1,815 tokens/s against SambaNova 707 and Groq 472 (benchmark table, undated): two partial alternatives, none at parity."},{"url":"https://www.sec.gov/Archives/edgar/data/0002021728/000162828026056186/cbrsannouncesfinancialresu.htm","date":"2026-08-12","support":"The company says OpenAI GPT-5.6 Sol runs at 750 tokens/s on Cerebras, and its wafer-scale design avoids HBM and CoWoS supply limits (Q2 2026 release, 12 Aug 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://huggingface.co/docs/inference-providers/index","date":"undated","support":"Hugging Face documentation (no date) lists Cerebras among 15 chat-model partners selectable through one OpenAI-style API."},{"url":"https://huggingface.co/blog/inference-providers","date":"2025-01-28","support":"Hugging Face says switching provider in its client needs only a changed provider name, with the rest of the code unchanged (posted 2025-01-28)."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"The nearest alternative, SambaNova, is about 2.6 times slower (707 against 1,815 tokens/s) and Groq about 3.8 times slower on gpt-oss-120b (benchmark table, undated)."},{"url":"https://www.sec.gov/Archives/edgar/data/0002021728/000162828026056186/cbrsannouncesfinancialresu.htm","date":"2026-08-12","support":"The company says capacity of over 600 MW is live or contracted and manufacturing will grow more than 10x in 2026 (12 Aug 2026); no rival capacity figure of that size was found."}]}}}}},"together-ai":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"together-ai","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":2,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Benchmark page (no publication date; 72-hour medians, read 2026-10-22) lists 16 API providers for gpt-oss-120b, including Baseten, DeepInfra, Nebius, Crusoe, Groq and Cerebras."},{"url":"https://fireworks.ai/blog/series-d-announcement","date":"2026-07-15","support":"The company (Fireworks) says it serves 40 trillion tokens a day and names Cursor, also a Together customer, as its own customer."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://docs.together.ai/docs/openai-api-compatibility","date":"undated","support":"Together docs (undated): API compatible with OpenAI REST and SDKs; switching needs two changes, API key and base URL."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://fireworks.ai/blog/series-d-announcement","date":"2026-07-15","support":"The company (Fireworks) says it passes $1bn annualised run-rate and 40 trillion tokens a day: a larger alternative exists."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"together-ai","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://huggingface.co/docs/inference-providers/index","date":"undated","support":"Hugging Face documentation (no date) lists 15 chat-model inference partners, including Baseten, Cerebras, DeepInfra, Fireworks, Groq, Novita and Together, behind one API."},{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Independent benchmark page (no date shown for the table) lists 16 API providers for gpt-oss-120b; the fastest are Cerebras 1,815, SambaNova 707 and Groq 472 tokens per second."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.together.ai/blog/announcing-our-series-c","date":"2026-07-01","support":"The company says Decagon cut inference costs sixfold after moving to it, an example of customers changing provider (1 Jul 2026)."},{"url":"https://docs.together.ai/docs/openai-api-compatibility","date":"undated","support":"Together's documentation (no date) says its API is compatible with the OpenAI REST API and SDKs: change the key and base URL."},{"url":"https://huggingface.co/blog/inference-providers","date":"2025-01-28","support":"Hugging Face says switching provider in its client needs only a changed provider name, with the rest of the code unchanged (posted 2025-01-28)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"On the same benchmark page (undated table) CoreWeave, DeepInfra and Novita are the cheapest and Cerebras the fastest, so no alternative is a step down from commodity open-model serving."},{"url":"https://fireworks.ai/blog/series-d-announcement","date":"2026-07-15","support":"Fireworks says it serves over 40 trillion tokens a day (15 Jul 2026), so a rival matches Together's scale (company statement)."}]}}}}},"modal":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"modal","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":2,"evidence":[{"url":"https://www.verdict.co.uk/baseten-secures-1-5bn-series-f/","date":"2026-06-23","support":"Verdict: Baseten runs dedicated custom-model inference across 87 clusters and 18 clouds."},{"url":"https://fireworks.ai/blog/series-d-announcement","date":"2026-07-15","support":"The company (Fireworks) says over 95% of its 40 trillion daily tokens come from customer-specialised models; several peers serve custom models."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://modal.com/docs/guide","date":"undated","support":"Modal docs (undated): containers and GPU choice are defined in Modal's Python SDK, so moving means porting deployment code; models stay standard (judgment: weeks)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://fireworks.ai/blog/series-d-announcement","date":"2026-07-15","support":"The company (Fireworks) says it passes $1bn run-rate; larger alternatives exist for custom-model serving."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"modal","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://huggingface.co/docs/inference-providers/index","date":"undated","support":"Hugging Face documentation (no date) lists 15 chat-model inference partners, including Baseten, Cerebras, DeepInfra, Fireworks, Groq, Novita and Together, behind one API."},{"url":"https://nebius.com/newsroom/nebius-launches-nebius-token-factory-to-deliver-production-ai-inference-at-scale","date":"2025-11-05","support":"Nebius says customers can host their own models and use dedicated endpoints and fine-tuning on Token Factory (5 Nov 2025), another route for custom model serving."},{"url":"https://baseten.co/blog/announcing-our-series-f/","date":"2026-06-22","support":"Baseten says it post-trains and serves specialised models for customers such as Cursor and Harvey (22 Jun 2026)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://modal.com/blog/1-billion-annualized-run-rate-series-d","date":"2026-10-14","support":"The company says customers run their own code on its runtime over a capacity pool spanning 20+ clouds (14 Oct 2026); moving that code to another platform means re-engineering (inference, thin evidence)."},{"url":"https://a16z.com/ai-enterprise-2025/","date":"2025-06-10","support":"a16z's CIO survey says switching costs are rising and changing models can take a lot of engineering time; buyers go direct or via Databricks when their main cloud lacks a model (10 Jun 2025)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://baseten.co/blog/announcing-our-series-f/","date":"2026-06-22","support":"Baseten says its revenue grew 20x and inference volume 40x in a year, serving specialised models for Cursor and Harvey (company statement, 22 Jun 2026): a comparable alternative for custom model serving."}]}}}}},"volcano-engine":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"volcano-engine","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"china-serving":{"alternatives":{"level":2,"evidence":[{"url":"https://www.yicaiglobal.com/news/tencent-bytedance-other-chinese-tech-giants-launch-deepseek-models-via-cloud-platforms","date":"2025-02-05","support":"Yicai: Baidu, Alibaba, Tencent Cloud and Volcano Engine all launched DeepSeek-R1/V3 deployment within days of each other (Feb 2025)."},{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC 2025 ranking by calls: Volcano Engine nearly half, then Alibaba Cloud, Baidu AI Cloud, SiliconFlow, China Mobile Cloud."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC: over the past two years price was almost always buyers' top selection factor for China MaaS; buyers usually call model APIs without secondary tuning."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://kr-asia.com/how-bytedances-volcano-engine-holds-nearly-half-of-chinas-maas-market","date":"2026-05-13","support":"KrASIA citing IDC: Volcano Engine held 49.5% of China's MaaS market in 2025; scale and engineering keep its prices below average, rivals face cost pressure (small gap)."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"volcano-engine","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"china-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC ranks China public-cloud MaaS by 2025 call volume: Volcano Engine, Alibaba Cloud, Baidu AI Cloud, SiliconFlow, China Mobile Cloud, and also names Tencent, SenseTime, Huawei and China Telecom (7 May 2026)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://winbuzzer.com/2026/09/09/deepseek-reportedly-plans-160000-plus-huawei-chips-model-serving-xcxwbn/","date":"2026-09-09","support":"Citing Bloomberg, Huawei could build only the low hundreds of thousands of Ascend 950DT chips in 2026 and a big order may take over a year, limiting how fast rivals can add capacity (inference)."},{"url":"https://www.alibabacloud.com/help/en/model-studio/compatibility-of-openai-with-dashscope","date":"2026-10-13","support":"Alibaba Cloud's documentation (updated Oct 13, 2026) says OpenAI code migrates by changing only key, base URL and model name, and lists DeepSeek, Kimi, GLM and MiniMax as supported."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC puts Volcano Engine at nearly half of 2025 public-cloud call volume and over 40% of revenue, ahead of Alibaba Cloud and the rest (7 May 2026)."},{"url":"https://technode.com/2026/04/07/doubao-surpasses-120-trillion-daily-tokens-as-usage-doubles-in-three-months/","date":"2026-04-07","support":"Volcano Engine's Doubao passed 120 trillion tokens a day in March 2026 (company statement; no split between own and third-party models)."}]}}}}},"oracle-cloud-infrastructure":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"oracle-cloud-infrastructure","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"cloud-model-platforms":{"alternatives":{"level":1,"evidence":[{"url":"https://docs.oracle.com/en-us/iaas/Content/generative-ai/pretrained-models.htm","date":"undated","support":"Oracle docs (undated): OCI serves xAI Grok, Cohere, Google Gemini 2.5, Meta Llama and OpenAI gpt-oss, families also sold on other major clouds."},{"url":"https://docs.databricks.com/aws/en/machine-learning/model-serving/foundation-model-overview","date":"2026-10-19","support":"Databricks docs (updated 19 Oct 2026) host Grok, Gemini, Llama, gpt-oss and others too; many platforms carry the same models."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://docs.oracle.com/en-us/iaas/Content/generative-ai/overview.htm","date":"undated","support":"Oracle docs (undated): customers can import, fine-tune and host custom models on dedicated AI clusters; moving them means re-hosting and new cloud approvals (evidence thin; judgment 3-12 months)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://aws.amazon.com/about-aws/whats-new/2026/06/amazon-bedrock-openai-models-codex-generally-available/","date":"2026-06-01","support":"AWS: Bedrock runs OpenAI frontier models in production alongside open models; larger platforms exceed OCI's model range."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"oracle-cloud-infrastructure","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"cloud-model-platforms":{"alternatives":{"level":1,"evidence":[{"url":"https://docs.aws.amazon.com/bedrock/latest/userguide/model-cards.html","date":"undated","support":"AWS's model page (no date) lists the same families as OCI's third-party list (DeepSeek, Qwen, Grok, NVIDIA Nemotron, Google Gemma, Meta, Mistral, Cohere)."},{"url":"https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/foundry-models-overview","date":"2026-07-28","support":"Foundry offers over 10,000 models (page metadata date 2026-07-28); with Bedrock, Google, Databricks and Snowflake that is six or more platforms."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://a16z.com/ai-enterprise-2025/","date":"2025-06-10","support":"a16z's CIO survey says switching costs are rising and changing models can take a lot of engineering time; buyers go direct or via Databricks when their main cloud lacks a model (10 Jun 2025)."},{"url":"https://docs.oracle.com/iaas/releasenotes/services/generative-ai/index.htm","date":"2026-10-20","support":"Oracle's release notes (entries to 20 Oct 2026) say most open-weight entries are imports the customer hosts itself, so moving them means redeploying."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://docs.oracle.com/iaas/releasenotes/services/generative-ai/index.htm","date":"2026-10-20","support":"Oracle's 2026 release-note entries name Qwen, NVIDIA, DeepSeek, xAI, Google and Cohere models but no Claude or OpenAI frontier model."},{"url":"https://docs.aws.amazon.com/bedrock/latest/userguide/model-cards.html","date":"undated","support":"AWS's model page (no date) carries every one of those families and more, so the alternatives lose nothing."}]}}}}},"groq":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"groq","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":2,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Benchmark page (undated; read 2026-10-22): Cerebras and SambaNova beat Groq's 472 t/s on gpt-oss-120b; Crusoe, Azure and Nebius reach about 370-382 t/s."},{"url":"https://groq.com/newsroom/groq-and-nvidia-enter-non-exclusive-inference-technology-licensing-agreement-to-accelerate-ai-inference-at-global-scale","date":"2025-12-24","support":"Groq: GroqCloud continues without interruption after the non-exclusive NVIDIA licence; founder and president join NVIDIA."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://console.groq.com/docs/openai","date":"undated","support":"Groq docs (undated): mostly compatible with OpenAI client libraries; switch by changing API key and base URL."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Benchmark page (undated): Cerebras (1,815 t/s) and SambaNova (707 t/s) exceed Groq on gpt-oss-120b, so the best alternative has no gap."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"groq","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":3,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Only Cerebras (1,815) and SambaNova (707) are faster than Groq (472 tokens/s) on gpt-oss-120b (benchmark table, undated); GPU hosts are slower, so few alternatives match its speed class."},{"url":"https://www.sec.gov/Archives/edgar/data/0002021728/000162828026056186/cbrsannouncesfinancialresu.htm","date":"2026-08-12","support":"Cerebras says GPT-5.6 Sol runs at 750 tokens/s on its cloud, whose revenue grew 281% in Q2 2026 (12 Aug 2026): a qualified fast alternative."},{"url":"https://www.pymnts.com/news/investment-tracker/2026/groq-seeks-650-million-amid-shift-to-ai-inference-neocloud-business/","date":"2026-05-28","support":"NVIDIA took a non-exclusive licence to Groq's IP and hired its founder in Dec 2025; Groq says GroqCloud continues independently (reported 28 May 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://huggingface.co/docs/inference-providers/index","date":"undated","support":"Hugging Face documentation (no date) lists Groq among 15 chat-model partners selectable through one API."},{"url":"https://huggingface.co/blog/inference-providers","date":"2025-01-28","support":"Hugging Face says switching provider in its client needs only a changed provider name, with the rest of the code unchanged (posted 2025-01-28)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Cerebras and SambaNova are faster than Groq on the same benchmark page (undated table), so the best alternative does not fall short in speed."}]}}}}},"databricks":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"databricks","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"cloud-model-platforms":{"alternatives":{"level":1,"evidence":[{"url":"https://docs.databricks.com/aws/en/machine-learning/model-serving/foundation-model-overview","date":"2026-10-19","support":"Databricks docs (updated 19 Oct 2026): hosted models come from Anthropic, OpenAI, Google, Meta, Qwen, xAI and others, all also sold by clouds and the developers."},{"url":"https://www.snowflake.com/en/news/press-releases/snowflake-and-anthropic-announce-200-million-partnership-to-bring-agentic-ai-to-global-enterprises/","date":"2025-12-03","support":"Snowflake: Claude reaches 12,600+ Snowflake customers across all three major clouds; a direct data-platform alternative."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://docs.databricks.com/aws/en/machine-learning/model-serving/foundation-model-overview","date":"2026-10-19","support":"Databricks docs: customers can also use Unity Gateway with their own model provider account, so traffic can move off Databricks-hosted models within weeks."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://blogs.microsoft.com/blog/2025/11/18/microsoft-nvidia-and-anthropic-announce-strategic-partnerships/","date":"2025-11-18","support":"Microsoft: Claude available on all three leading clouds; the same models run at larger scale elsewhere."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"databricks","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"cloud-model-platforms":{"alternatives":{"level":2,"evidence":[{"url":"https://docs.snowflake.com/user-guide/snowflake-cortex/aisql-regional-availability","date":"undated","support":"Snowflake's documentation (no date) lists Claude, OpenAI GPT, Gemini, Llama and Mistral in Cortex AI Functions, the nearest like-for-like data platform."},{"url":"https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/foundry-models-overview","date":"2026-07-28","support":"Foundry lists over 10,000 models, including Anthropic, DeepSeek, Meta and Mistral (page metadata date 2026-07-28)."},{"url":"https://docs.aws.amazon.com/bedrock/latest/userguide/model-cards.html","date":"undated","support":"AWS's model page (no date) lists Claude 5.x and OpenAI GPT-6 among 19 providers."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://a16z.com/ai-enterprise-2025/","date":"2025-06-10","support":"a16z's CIO survey says switching costs are rising and changing models can take a lot of engineering time; buyers go direct or via Databricks when their main cloud lacks a model (10 Jun 2025)."},{"url":"https://blogs.microsoft.com/blog/2025/11/18/microsoft-nvidia-and-anthropic-announce-strategic-partnerships/","date":"2025-11-18","support":"Microsoft announced Claude for Foundry customers on 2025-11-18, with Anthropic committing $30bn of Azure compute."},{"url":"https://azure.microsoft.com/en-us/blog/claude-in-microsoft-foundry-is-now-generally-available/","date":"2026-06-29","support":"Azure's blog dates general availability of Claude on Foundry to 2026-06-29, about seven months after the announcement: a rough guide to how long a platform needs to qualify and ramp a model line."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://docs.databricks.com/aws/en/machine-learning/model-serving/foundation-model-overview","date":"2026-10-19","support":"Databricks lists Claude Opus 5, Sonnet 5.5 and Fable 5.1, GPT-6 Astra, Gemini 4 and Llama 4 (updated Oct 19, 2026)."},{"url":"https://docs.snowflake.com/user-guide/snowflake-cortex/aisql-regional-availability","date":"undated","support":"Snowflake's page (no date) calls claude-opus-5 and gemini-3.1-pro its most capable models, a step behind Databricks' list."}]}}}}},"snowflake":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"snowflake","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"cloud-model-platforms":{"alternatives":{"level":1,"evidence":[{"url":"https://docs.snowflake.com/en/user-guide/snowflake-cortex/aisql","date":"undated","support":"Snowflake docs (undated): Cortex AI offers OpenAI, Anthropic, Meta, Mistral, DeepSeek and xAI models, all also offered by clouds and the developers."},{"url":"https://docs.databricks.com/aws/en/machine-learning/model-serving/foundation-model-overview","date":"2026-10-19","support":"Databricks docs (updated 19 Oct 2026) host the same model families; many qualified alternatives."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://docs.snowflake.com/en/user-guide/snowflake-cortex/aisql","date":"undated","support":"Snowflake docs: LLMs are deployed within the Snowflake Service perimeter; moving off means new data-egress and governance approvals (evidence thin; judgment 3-12 months)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.snowflake.com/en/news/press-releases/snowflake-and-anthropic-announce-200-million-partnership-to-bring-agentic-ai-to-global-enterprises/","date":"2025-12-03","support":"Snowflake: thousands of customers process trillions of Claude tokens a month in Cortex; the same Claude models are served at larger scale by Bedrock, Vertex and Foundry."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"snowflake","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"cloud-model-platforms":{"alternatives":{"level":2,"evidence":[{"url":"https://docs.databricks.com/aws/en/machine-learning/model-serving/foundation-model-overview","date":"2026-10-19","support":"Databricks hosts Claude 5.x, GPT-6, Gemini and Llama models (updated Oct 19, 2026): the closest like-for-like data platform."},{"url":"https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/foundry-models-overview","date":"2026-07-28","support":"Hyperscaler platforms such as Foundry (over 10,000 models, page metadata date 2026-07-28) serve the same Claude and OpenAI lines."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://a16z.com/ai-enterprise-2025/","date":"2025-06-10","support":"a16z's CIO survey says switching costs are rising and changing models can take a lot of engineering time; buyers go direct or via Databricks when their main cloud lacks a model (10 Jun 2025)."},{"url":"https://blogs.microsoft.com/blog/2025/11/18/microsoft-nvidia-and-anthropic-announce-strategic-partnerships/","date":"2025-11-18","support":"Microsoft announced Claude for Foundry customers on 2025-11-18, with Anthropic committing $30bn of Azure compute."},{"url":"https://azure.microsoft.com/en-us/blog/claude-in-microsoft-foundry-is-now-generally-available/","date":"2026-06-29","support":"Azure's blog dates general availability of Claude on Foundry to 2026-06-29, about seven months after the announcement: a rough guide to how long a platform needs to qualify and ramp a model line."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://docs.databricks.com/aws/en/machine-learning/model-serving/foundation-model-overview","date":"2026-10-19","support":"Databricks lists newer Claude 5.5 and Fable 5.1 models plus GPT-6 and Gemini 4 (updated Oct 19, 2026)."},{"url":"https://docs.snowflake.com/user-guide/snowflake-cortex/aisql-regional-availability","date":"undated","support":"Snowflake's page (no date) calls claude-opus-5 and gemini-3.1-pro its most capable models, so rivals offer at least as much."}]}}}}},"alibaba-cloud":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"alibaba-cloud","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"china-serving":{"alternatives":{"level":2,"evidence":[{"url":"https://www.yicaiglobal.com/news/tencent-bytedance-other-chinese-tech-giants-launch-deepseek-models-via-cloud-platforms","date":"2025-02-05","support":"Yicai: Baidu, Alibaba, Tencent Cloud and Volcano Engine all launched DeepSeek-R1/V3 deployment within days of each other (Feb 2025)."},{"url":"https://www.news.cn/finance/20250213/1d369f19be5f42c29b150bc628f8ed05/c.html","date":"2025-02-13","support":"Xinhua/MIIT: all three basic telecom operators fully integrated DeepSeek open models; China Telecom and China Mobile on domestic Ascend stacks."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC: over the past two years price was almost always buyers' top selection factor for China MaaS; buyers usually call model APIs without secondary tuning."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC 2025 ranking by calls: Volcano Engine nearly half, then Alibaba Cloud, Baidu AI Cloud, SiliconFlow, China Mobile Cloud."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"alibaba-cloud","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"china-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC ranks China public-cloud MaaS by 2025 call volume: Volcano Engine, Alibaba Cloud, Baidu AI Cloud, SiliconFlow, China Mobile Cloud, and also names Tencent, SenseTime, Huawei and China Telecom (7 May 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.alibabacloud.com/help/en/model-studio/compatibility-of-openai-with-dashscope","date":"2026-10-13","support":"Alibaba Cloud's documentation (updated Oct 13, 2026) says OpenAI code migrates by changing only key, base URL and model name, and lists DeepSeek, Kimi, GLM and MiniMax as supported."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC lists Volcano Engine ahead of Alibaba Cloud by 2025 call volume and revenue (7 May 2026), so the best alternative is not a step down."}]}}}}},"deepinfra":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"deepinfra","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Benchmark page (no publication date; 72-hour medians, read 2026-10-22) lists 16 API providers for gpt-oss-120b, including DeepInfra itself (52 t/s) and faster providers."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://docs.deepinfra.com/chat/overview","date":"undated","support":"DeepInfra docs (undated): OpenAI-compatible chat completions; switch by changing base URL, key and model name."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://fireworks.ai/blog/series-d-announcement","date":"2026-07-15","support":"The company (Fireworks) says it serves over 40 trillion tokens a day; much larger providers can absorb this volume."},{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Benchmark page (undated): the same open models run at similar or higher speed at many providers."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"deepinfra","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://huggingface.co/docs/inference-providers/index","date":"undated","support":"Hugging Face documentation (no date) lists 15 chat-model inference partners, including Baseten, Cerebras, DeepInfra, Fireworks, Groq, Novita and Together, behind one API."},{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Independent benchmark page (no date shown for the table) lists 16 API providers for gpt-oss-120b; the fastest are Cerebras 1,815, SambaNova 707 and Groq 472 tokens per second."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://deepinfra.com/blog/deepinfra-series-b","date":"2026-05-04","support":"The company says it offers 150+ open-source models through OpenAI-compatible APIs from eight US data centres (4 May 2026)."},{"url":"https://huggingface.co/blog/inference-providers","date":"2025-01-28","support":"Hugging Face says switching provider in its client needs only a changed provider name, with the rest of the code unchanged (posted 2025-01-28)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"On the same benchmark page (undated table) CoreWeave, DeepInfra and Novita are the cheapest and Cerebras the fastest, so no alternative is a step down from commodity open-model serving."}]}}}}},"digitalocean":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"digitalocean","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Benchmark page (no publication date; 72-hour medians, read 2026-10-22) lists 16 API providers for gpt-oss-120b, including Together AI, DeepInfra, Nebius and the hyperscalers."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://docs.digitalocean.com/products/gradient-ai-platform/how-to/use-serverless-inference/","date":"2026-07-13","support":"DigitalOcean docs (page last verified 13 Jul 2026): serverless inference uses Chat Completions, Responses and Messages APIs, standard interfaces."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://fireworks.ai/blog/series-d-announcement","date":"2026-07-15","support":"The company (Fireworks) says it serves over 40 trillion tokens a day; much larger providers can absorb this volume."},{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Benchmark page (undated): the same open models run at similar or higher speed at many providers."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"digitalocean","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://huggingface.co/docs/inference-providers/index","date":"undated","support":"Hugging Face documentation (no date) lists 15 chat-model inference partners, including Baseten, Cerebras, DeepInfra, Fireworks, Groq, Novita and Together, behind one API."},{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Independent benchmark page (no date shown for the table) lists 16 API providers for gpt-oss-120b; the fastest are Cerebras 1,815, SambaNova 707 and Groq 472 tokens per second."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://docs.digitalocean.com/products/gradient-ai-platform/how-to/use-serverless-inference/","date":"2026-07-13","support":"DigitalOcean's documentation (last verified 13 Jul 2026) lists a model catalog of hosted and third-party models with Chat Completions and Responses API guides."},{"url":"https://huggingface.co/blog/inference-providers","date":"2025-01-28","support":"Hugging Face says switching provider in its client needs only a changed provider name, with the rest of the code unchanged (posted 2025-01-28)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"On the same benchmark page (undated table) CoreWeave, DeepInfra and Novita are the cheapest and Cerebras the fastest, so no alternative is a step down from commodity open-model serving."}]}}}}},"baidu-ai-cloud":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"baidu-ai-cloud","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"china-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://www.yicaiglobal.com/news/tencent-bytedance-other-chinese-tech-giants-launch-deepseek-models-via-cloud-platforms","date":"2025-02-05","support":"Yicai: Baidu, Alibaba, Tencent Cloud and Volcano Engine all launched DeepSeek-R1/V3 deployment within days of each other (Feb 2025)."},{"url":"https://www.news.cn/finance/20250213/1d369f19be5f42c29b150bc628f8ed05/c.html","date":"2025-02-13","support":"Xinhua/MIIT: all three basic telecom operators fully integrated DeepSeek open models; China Telecom and China Mobile on domestic Ascend stacks."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC: over the past two years price was almost always buyers' top selection factor for China MaaS; buyers usually call model APIs without secondary tuning."},{"url":"https://www.yicaiglobal.com/news/tencent-bytedance-other-chinese-tech-giants-launch-deepseek-models-via-cloud-platforms","date":"2025-02-05","support":"Yicai: Baidu priced its DeepSeek-R1 API at half DeepSeek's official price, a price-led market."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC 2025 ranking by calls: Volcano Engine nearly half, then Alibaba Cloud, Baidu AI Cloud, SiliconFlow, China Mobile Cloud."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"baidu-ai-cloud","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"china-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC ranks China public-cloud MaaS by 2025 call volume: Volcano Engine, Alibaba Cloud, Baidu AI Cloud, SiliconFlow, China Mobile Cloud, and also names Tencent, SenseTime, Huawei and China Telecom (7 May 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.alibabacloud.com/help/en/model-studio/compatibility-of-openai-with-dashscope","date":"2026-10-13","support":"Alibaba Cloud's documentation (updated Oct 13, 2026) says OpenAI code migrates by changing only key, base URL and model name, and lists DeepSeek, Kimi, GLM and MiniMax as supported."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC ranks Volcano Engine and Alibaba Cloud ahead of Baidu AI Cloud by 2025 call volume (7 May 2026)."},{"url":"https://www.sec.gov/Archives/edgar/data/0001329099/000119312526355431/d112854dex991.htm","date":"2026-08-18","support":"Baidu's Q2 2026 release shows AI cloud infrastructure revenue of RMB 7.3bn, up 50% (company figure), a scale peers match."}]}}}}},"china-mobile-cloud":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"china-mobile-cloud","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"china-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://www.yicaiglobal.com/news/tencent-bytedance-other-chinese-tech-giants-launch-deepseek-models-via-cloud-platforms","date":"2025-02-05","support":"Yicai: Baidu, Alibaba, Tencent Cloud and Volcano Engine all launched DeepSeek-R1/V3 deployment within days of each other (Feb 2025)."},{"url":"https://www.news.cn/finance/20250213/1d369f19be5f42c29b150bc628f8ed05/c.html","date":"2025-02-13","support":"Xinhua/MIIT: all three basic telecom operators fully integrated DeepSeek open models; China Telecom and China Mobile on domestic Ascend stacks."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC: over the past two years price was almost always buyers' top selection factor for China MaaS; buyers usually call model APIs without secondary tuning."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC 2025 ranking by calls: Volcano Engine nearly half, then Alibaba Cloud, Baidu AI Cloud, SiliconFlow, China Mobile Cloud."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"china-mobile-cloud","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"china-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC ranks China public-cloud MaaS by 2025 call volume: Volcano Engine, Alibaba Cloud, Baidu AI Cloud, SiliconFlow, China Mobile Cloud, and also names Tencent, SenseTime, Huawei and China Telecom (7 May 2026)."},{"url":"https://www.rcrwireless.com/20250212/asia-pacific/chinese-telcos-deepseek","date":"2025-02-12","support":"RCR Wireless reports China Mobile, China Telecom and China Unicom all integrated DeepSeek models on their clouds (12 Feb 2025)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.alibabacloud.com/help/en/model-studio/compatibility-of-openai-with-dashscope","date":"2026-10-13","support":"Alibaba Cloud's documentation (updated Oct 13, 2026) says OpenAI code migrates by changing only key, base URL and model name, and lists DeepSeek, Kimi, GLM and MiniMax as supported."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC puts China Mobile Cloud fifth by 2025 call volume, behind Volcano Engine, Alibaba, Baidu and SiliconFlow (7 May 2026)."}]}}}}},"huawei-cloud":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"huawei-cloud","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"china-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://www.yicaiglobal.com/news/tencent-bytedance-other-chinese-tech-giants-launch-deepseek-models-via-cloud-platforms","date":"2025-02-05","support":"Yicai: Baidu, Alibaba, Tencent Cloud and Volcano Engine all launched DeepSeek-R1/V3 deployment within days of each other (Feb 2025)."},{"url":"https://www.news.cn/finance/20250213/1d369f19be5f42c29b150bc628f8ed05/c.html","date":"2025-02-13","support":"Xinhua/MIIT: all three basic telecom operators fully integrated DeepSeek open models; China Telecom and China Mobile on domestic Ascend stacks."},{"url":"https://m.cls.cn/detail/1933192","date":"2025-02-01","support":"CLS: SiliconFlow and Huawei Cloud jointly launched DeepSeek R1/V3 inference on Huawei Cloud's Ascend service."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC: over the past two years price was almost always buyers' top selection factor for China MaaS; buyers usually call model APIs without secondary tuning."},{"url":"https://www.news.cn/finance/20250213/1d369f19be5f42c29b150bc628f8ed05/c.html","date":"2025-02-13","support":"Xinhua: China Telecom and China Mobile also serve DeepSeek on domestic Ascend stacks, so a domestic-chip requirement can be met elsewhere."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC 2025 ranking by calls: Volcano Engine nearly half, then Alibaba Cloud, Baidu AI Cloud, SiliconFlow, China Mobile Cloud."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"huawei-cloud","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"china-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC ranks China public-cloud MaaS by 2025 call volume: Volcano Engine, Alibaba Cloud, Baidu AI Cloud, SiliconFlow, China Mobile Cloud, and also names Tencent, SenseTime, Huawei and China Telecom (7 May 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.alibabacloud.com/help/en/model-studio/compatibility-of-openai-with-dashscope","date":"2026-10-13","support":"Alibaba Cloud's documentation (updated Oct 13, 2026) says OpenAI code migrates by changing only key, base URL and model name, and lists DeepSeek, Kimi, GLM and MiniMax as supported."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://arxiv.org/abs/2506.12708","date":"2025-06-15","support":"A June 2025 paper reports serving DeepSeek-R1 on Huawei CloudMatrix384 at 1,943 tokens/s per NPU in decode (authors' claim): peers serve the same open models on other chips."},{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC names Huawei Cloud outside the top five by call volume (7 May 2026)."}]}}}}},"tencent-cloud":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"tencent-cloud","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"china-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://www.yicaiglobal.com/news/tencent-bytedance-other-chinese-tech-giants-launch-deepseek-models-via-cloud-platforms","date":"2025-02-05","support":"Yicai: Baidu, Alibaba, Tencent Cloud and Volcano Engine all launched DeepSeek-R1/V3 deployment within days of each other (Feb 2025)."},{"url":"https://www.news.cn/finance/20250213/1d369f19be5f42c29b150bc628f8ed05/c.html","date":"2025-02-13","support":"Xinhua/MIIT: all three basic telecom operators fully integrated DeepSeek open models; China Telecom and China Mobile on domestic Ascend stacks."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC: over the past two years price was almost always buyers' top selection factor for China MaaS; buyers usually call model APIs without secondary tuning."},{"url":"https://www.yicaiglobal.com/news/tencent-bytedance-other-chinese-tech-giants-launch-deepseek-models-via-cloud-platforms","date":"2025-02-05","support":"Yicai: Tencent Cloud TI Platform launched one-click deployment of the full DeepSeek range."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC 2025 ranking by calls: Volcano Engine nearly half, then Alibaba Cloud, Baidu AI Cloud, SiliconFlow, China Mobile Cloud."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"tencent-cloud","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"china-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC ranks China public-cloud MaaS by 2025 call volume: Volcano Engine, Alibaba Cloud, Baidu AI Cloud, SiliconFlow, China Mobile Cloud, and also names Tencent, SenseTime, Huawei and China Telecom (7 May 2026)."},{"url":"https://www.tencentcloud.com/products/tokenhub","date":"undated","support":"Tencent Cloud's TokenHub page (no date) lists DeepSeek, Kimi, GLM, MiniMax and Qwen as third-party models, the same open models its rivals host."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.alibabacloud.com/help/en/model-studio/compatibility-of-openai-with-dashscope","date":"2026-10-13","support":"Alibaba Cloud's documentation (updated Oct 13, 2026) says OpenAI code migrates by changing only key, base URL and model name, and lists DeepSeek, Kimi, GLM and MiniMax as supported."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC ranks Volcano Engine and Alibaba Cloud well ahead of Tencent Cloud (7 May 2026), so the best alternative is not a step down."}]}}}}},"coreweave":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"coreweave","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Benchmark page (no publication date; 72-hour medians, read 2026-10-22) lists 16 API providers for gpt-oss-120b, including CoreWeave, Nebius, Crusoe and Together AI."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://docs.coreweave.com/products/inference/serverless.md","date":"undated","support":"CoreWeave docs (undated): serverless inference offers open-source models through an OpenAI-compatible API; switch by base URL and key."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://fireworks.ai/blog/series-d-announcement","date":"2026-07-15","support":"The company (Fireworks) says it serves over 40 trillion tokens a day; much larger providers can absorb this volume."},{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Benchmark page (undated): the same open models run at similar or higher speed at many providers."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"coreweave","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://huggingface.co/docs/inference-providers/index","date":"undated","support":"Hugging Face documentation (no date) lists 15 chat-model inference partners, including Baseten, Cerebras, DeepInfra, Fireworks, Groq, Novita and Together, behind one API."},{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Independent benchmark page (no date shown for the table) lists 16 API providers for gpt-oss-120b; the fastest are Cerebras 1,815, SambaNova 707 and Groq 472 tokens per second."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://docs.coreweave.com/products/inference/serverless.md","date":"undated","support":"CoreWeave's documentation (no date) describes serverless inference as an OpenAI-compatible API reached by changing the client's base URL."},{"url":"https://huggingface.co/blog/inference-providers","date":"2025-01-28","support":"Hugging Face says switching provider in its client needs only a changed provider name, with the rest of the code unchanged (posted 2025-01-28)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"CoreWeave is the cheapest of 16 providers on gpt-oss-120b (undated table); rivals such as Cerebras are faster, so alternatives are not a step down."}]}}}}},"crusoe":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"crusoe","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Benchmark page (no publication date; 72-hour medians, read 2026-10-22) lists 16 API providers for gpt-oss-120b, including Crusoe, Nebius, CoreWeave and Baseten."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://docs.vllm.ai/en/latest/deployment/frameworks/crusoe/","date":"undated","support":"vLLM docs (unlabelled page date 10 Aug 2026): Crusoe Managed Inference is an OpenAI-compatible, vLLM-based API; self-hosted vLLM code works unchanged."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://fireworks.ai/blog/series-d-announcement","date":"2026-07-15","support":"The company (Fireworks) says it serves over 40 trillion tokens a day; much larger providers can absorb this volume."},{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Benchmark page (undated): the same open models run at similar or higher speed at many providers."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"crusoe","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://huggingface.co/docs/inference-providers/index","date":"undated","support":"Hugging Face documentation (no date) lists 15 chat-model inference partners, including Baseten, Cerebras, DeepInfra, Fireworks, Groq, Novita and Together, behind one API."},{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Independent benchmark page (no date shown for the table) lists 16 API providers for gpt-oss-120b; the fastest are Cerebras 1,815, SambaNova 707 and Groq 472 tokens per second."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://docs.vllm.ai/en/latest/deployment/frameworks/crusoe/","date":"undated","support":"vLLM documentation (no date) says Crusoe Managed Inference is an OpenAI-compatible API and code for a self-hosted vLLM server works unchanged."},{"url":"https://huggingface.co/blog/inference-providers","date":"2025-01-28","support":"Hugging Face says switching provider in its client needs only a changed provider name, with the rest of the code unchanged (posted 2025-01-28)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"On the same benchmark page (undated table) CoreWeave, DeepInfra and Novita are the cheapest and Cerebras the fastest, so no alternative is a step down from commodity open-model serving."}]}}}}},"cloudflare-workers-ai":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"cloudflare-workers-ai","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Benchmark page (no publication date; 72-hour medians, read 2026-10-22) lists 16 API providers for gpt-oss-120b, including Together AI, DeepInfra, Nebius, Groq and Cerebras."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://developers.cloudflare.com/workers-ai/configuration/open-ai-compatibility/","date":"2026-09-18","support":"Cloudflare docs (updated 18 Sep 2026): Workers AI is called by swapping the baseURL in the standard OpenAI SDK."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://fireworks.ai/blog/series-d-announcement","date":"2026-07-15","support":"The company (Fireworks) says it serves over 40 trillion tokens a day; much larger providers exist."}]}},"media-model-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://techcrunch.com/2025/12/11/runware-raises-50m-series-a-from-dawn-capital-comcast-ventures-to-become-the-api-for-all-ai","date":"2025-12-11","support":"TechCrunch: Runware supports over 400,000 models for 200,000+ developers, competing with fal and Replicate."},{"url":"https://fal.ai/","date":"undated","support":"fal homepage (undated): 1,500+ image, video, audio and 3D models in production."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://blog.cloudflare.com/replicate-joins-cloudflare/","date":"2025-11-17","support":"Cloudflare: Replicate's 50,000+ models are open-source and fine-tuned models; APIs continue to work, so moving them is a standard API change (judgment: weeks)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2025/12/09/fal-nabs-140m-in-fresh-funding-led-by-sequoia-tripling-valuation-to-4-5b/","date":"2025-12-09","support":"TechCrunch citing Bloomberg: fal passed $200m revenue by October 2025; a larger media host exists."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"cloudflare-workers-ai","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://huggingface.co/docs/inference-providers/index","date":"undated","support":"Hugging Face documentation (no date) lists 15 chat-model inference partners, including Baseten, Cerebras, DeepInfra, Fireworks, Groq, Novita and Together, behind one API."},{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Independent benchmark page (no date shown for the table) lists 16 API providers for gpt-oss-120b; the fastest are Cerebras 1,815, SambaNova 707 and Groq 472 tokens per second."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://developers.cloudflare.com/workers-ai/configuration/open-ai-compatibility/","date":"2026-09-18","support":"Cloudflare's documentation (last updated 18 Sep 2026) says Workers AI offers OpenAI-compatible chat and embeddings endpoints reached by changing the base URL."},{"url":"https://huggingface.co/blog/inference-providers","date":"2025-01-28","support":"Hugging Face says switching provider in its client needs only a changed provider name, with the rest of the code unchanged (posted 2025-01-28)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"On the same benchmark page (undated table) CoreWeave, DeepInfra and Novita are the cheapest and Cerebras the fastest, so no alternative is a step down from commodity open-model serving."}]}},"media-model-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://huggingface.co/docs/inference-providers/index","date":"undated","support":"Hugging Face documentation (no date) lists six text-to-image and five text-to-video partners, so media hosting has many suppliers."},{"url":"https://www.cityam.com/runware-raises-50m-to-accelerate-ai-media-generation/","date":"2025-12-12","support":"Runware says it serves 200,000 developers and counts Wix and Quora as customers (12 Dec 2025)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://huggingface.co/blog/inference-providers","date":"2025-01-28","support":"Hugging Face's launch providers included fal and Replicate, and switching needed only a changed provider name (28 Jan 2025)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://tech.eu/2025/12/09/fal-raises-140m-series-d-to-power-the-next-era-of-real-time-generative-media/","date":"2025-12-09","support":"fal, the largest media host in this register, says it serves millions of developers (9 Dec 2025), so the best alternative is not a step down (company statement)."}]}}}}},"hugging-face":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"hugging-face","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Benchmark page (no publication date; 72-hour medians, read 2026-10-22) lists 16 API providers for gpt-oss-120b, including Baseten, Together AI, Nebius and the hyperscalers."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://huggingface.co/docs/inference-endpoints/index","date":"undated","support":"Hugging Face docs (undated): Inference Endpoints deploy models with open engines (vLLM, TGI) or custom containers, which run elsewhere."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://fireworks.ai/blog/series-d-announcement","date":"2026-07-15","support":"The company (Fireworks) says it serves over 40 trillion tokens a day; much larger providers can absorb this volume."},{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Benchmark page (undated): the same open models run at similar or higher speed at many providers."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"hugging-face","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://huggingface.co/docs/inference-providers/index","date":"undated","support":"Hugging Face's own documentation (no date) shows its router passes requests to 15 chat-model partners at the partners' rates with no markup, so customers can go direct to the same hosts or to other routers."},{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Independent benchmark page (no date shown for the table) lists 16 API providers for gpt-oss-120b; the fastest are Cerebras 1,815, SambaNova 707 and Groq 472 tokens per second."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://huggingface.co/blog/inference-providers","date":"2025-01-28","support":"Hugging Face says switching provider in its client needs only a changed provider name, with the rest of the code unchanged (posted 2025-01-28)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://huggingface.co/docs/inference-providers/index","date":"undated","support":"Hugging Face documentation (no date) says the router adds no markup and the same providers are reachable directly, so alternatives lose nothing."}]}}}}},"nebius":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"nebius","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Benchmark page (no publication date; 72-hour medians, read 2026-10-22) lists 16 API providers for gpt-oss-120b, including Nebius, CoreWeave, Crusoe and Together AI."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://docs.tokenfactory.nebius.com/","date":"undated","support":"Nebius docs (undated): Token Factory exposes an OpenAI-compatible API; examples use the standard OpenAI SDK."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://fireworks.ai/blog/series-d-announcement","date":"2026-07-15","support":"The company (Fireworks) says it serves over 40 trillion tokens a day; much larger providers can absorb this volume."},{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Benchmark page (undated): the same open models run at similar or higher speed at many providers."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"nebius","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"independent-llm-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://huggingface.co/docs/inference-providers/index","date":"undated","support":"Hugging Face documentation (no date) lists 15 chat-model inference partners, including Baseten, Cerebras, DeepInfra, Fireworks, Groq, Novita and Together, behind one API."},{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"Independent benchmark page (no date shown for the table) lists 16 API providers for gpt-oss-120b; the fastest are Cerebras 1,815, SambaNova 707 and Groq 472 tokens per second."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://nebius.com/newsroom/nebius-launches-nebius-token-factory-to-deliver-production-ai-inference-at-scale","date":"2025-11-05","support":"The company says Token Factory offers OpenAI-compatible APIs to ease migration from other endpoints and serves 60+ open models (5 Nov 2025)."},{"url":"https://huggingface.co/blog/inference-providers","date":"2025-01-28","support":"Hugging Face says switching provider in its client needs only a changed provider name, with the rest of the code unchanged (posted 2025-01-28)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://artificialanalysis.ai/models/gpt-oss-120b/providers","date":"undated","support":"On the same benchmark page (undated table) CoreWeave, DeepInfra and Novita are the cheapest and Cerebras the fastest, so no alternative is a step down from commodity open-model serving."}]}}}}},"ppio":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"ppio","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"china-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://www.yicaiglobal.com/news/tencent-bytedance-other-chinese-tech-giants-launch-deepseek-models-via-cloud-platforms","date":"2025-02-05","support":"Yicai: Baidu, Alibaba, Tencent Cloud and Volcano Engine all launched DeepSeek-R1/V3 deployment within days of each other (Feb 2025)."},{"url":"https://www.news.cn/finance/20250213/1d369f19be5f42c29b150bc628f8ed05/c.html","date":"2025-02-13","support":"Xinhua/MIIT: all three basic telecom operators fully integrated DeepSeek open models; China Telecom and China Mobile on domestic Ascend stacks."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC: over the past two years price was almost always buyers' top selection factor for China MaaS; buyers usually call model APIs without secondary tuning."},{"url":"https://m.tech.china.com/articles/20250501/202505011668457.html","date":"2025-05-01","support":"China.com: PPIO serves DeepSeek R1/V3, Llama, GLM and Qwen, callable with one line of code."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC 2025 ranking by calls: Volcano Engine nearly half, then Alibaba Cloud, Baidu AI Cloud, SiliconFlow, China Mobile Cloud."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"ppio","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"china-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC ranks China public-cloud MaaS by 2025 call volume: Volcano Engine, Alibaba Cloud, Baidu AI Cloud, SiliconFlow, China Mobile Cloud, and also names Tencent, SenseTime, Huawei and China Telecom (7 May 2026)."},{"url":"https://eu.36kr.com/en/p/3901197105071748","date":"2026-07-18","support":"36Kr reports PPIO ranked first among independent AI clouds in China by CIC data, with 1.03 trillion tokens a day in April 2026 (18 Jul 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.alibabacloud.com/help/en/model-studio/compatibility-of-openai-with-dashscope","date":"2026-10-13","support":"Alibaba Cloud's documentation (updated Oct 13, 2026) says OpenAI code migrates by changing only key, base URL and model name, and lists DeepSeek, Kimi, GLM and MiniMax as supported."},{"url":"https://eu.36kr.com/en/p/3901197105071748","date":"2026-07-18","support":"PPIO schedules over 5,000 distributed compute nodes across six continents (36Kr, 18 Jul 2026); moving to larger Chinese clouds is plausible (inference)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC's ranking puts Volcano Engine and Alibaba Cloud ahead by call volume (7 May 2026), so larger alternatives can absorb PPIO's volume."}]}}}}},"ibm":{},"china-telecom-cloud":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"china-telecom-cloud","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"china-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://www.yicaiglobal.com/news/tencent-bytedance-other-chinese-tech-giants-launch-deepseek-models-via-cloud-platforms","date":"2025-02-05","support":"Yicai: Baidu, Alibaba, Tencent Cloud and Volcano Engine all launched DeepSeek-R1/V3 deployment within days of each other (Feb 2025)."},{"url":"https://www.news.cn/finance/20250213/1d369f19be5f42c29b150bc628f8ed05/c.html","date":"2025-02-13","support":"Xinhua/MIIT: all three basic telecom operators fully integrated DeepSeek open models; China Telecom and China Mobile on domestic Ascend stacks."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC: over the past two years price was almost always buyers' top selection factor for China MaaS; buyers usually call model APIs without secondary tuning."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC 2025 ranking by calls: Volcano Engine nearly half, then Alibaba Cloud, Baidu AI Cloud, SiliconFlow, China Mobile Cloud."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"china-telecom-cloud","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"china-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC ranks China public-cloud MaaS by 2025 call volume: Volcano Engine, Alibaba Cloud, Baidu AI Cloud, SiliconFlow, China Mobile Cloud, and also names Tencent, SenseTime, Huawei and China Telecom (7 May 2026)."},{"url":"https://www.rcrwireless.com/20250212/asia-pacific/chinese-telcos-deepseek","date":"2025-02-12","support":"RCR Wireless reports China Telecom was first of the big telcos to run full-stack DeepSeek-R1/V3 inference on its platform (12 Feb 2025)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.alibabacloud.com/help/en/model-studio/compatibility-of-openai-with-dashscope","date":"2026-10-13","support":"Alibaba Cloud's documentation (updated Oct 13, 2026) says OpenAI code migrates by changing only key, base URL and model name, and lists DeepSeek, Kimi, GLM and MiniMax as supported."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC names China Telecom Cloud outside the top five by call volume (7 May 2026), so larger rivals are not a step down."}]}}}}},"runpod":{},"sambanova":{},"sap":{},"siliconflow":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"siliconflow","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"china-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://www.yicaiglobal.com/news/tencent-bytedance-other-chinese-tech-giants-launch-deepseek-models-via-cloud-platforms","date":"2025-02-05","support":"Yicai: Baidu, Alibaba, Tencent Cloud and Volcano Engine all launched DeepSeek-R1/V3 deployment within days of each other (Feb 2025)."},{"url":"https://www.news.cn/finance/20250213/1d369f19be5f42c29b150bc628f8ed05/c.html","date":"2025-02-13","support":"Xinhua/MIIT: all three basic telecom operators fully integrated DeepSeek open models; China Telecom and China Mobile on domestic Ascend stacks."},{"url":"https://m.cls.cn/detail/1933192","date":"2025-02-01","support":"CLS: SiliconFlow and Huawei Cloud jointly launched DeepSeek R1/V3 inference on Huawei Cloud's Ascend service."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC: over the past two years price was almost always buyers' top selection factor for China MaaS; buyers usually call model APIs without secondary tuning."},{"url":"https://www.siliconflow.com/","date":"undated","support":"SiliconFlow homepage (undated): fully OpenAI-compatible API to 200+ open and commercial models (DeepSeek, Qwen, Kimi, GLM)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC 2025 ranking by calls: Volcano Engine nearly half, then Alibaba Cloud, Baidu AI Cloud, SiliconFlow, China Mobile Cloud."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"siliconflow","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"china-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC ranks China public-cloud MaaS by 2025 call volume: Volcano Engine, Alibaba Cloud, Baidu AI Cloud, SiliconFlow, China Mobile Cloud, and also names Tencent, SenseTime, Huawei and China Telecom (7 May 2026)."},{"url":"https://www.caixinglobal.com/2026-06-17/siliconflow-raises-294-million-as-chinas-ai-inference-demand-surges-102454833.html","date":"2026-06-17","support":"Caixin reports SiliconFlow ranked fourth in China's MaaS market by 2025 token use, behind ByteDance and Alibaba Cloud (17 Jun 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.alibabacloud.com/help/en/model-studio/compatibility-of-openai-with-dashscope","date":"2026-10-13","support":"Alibaba Cloud's documentation (updated Oct 13, 2026) says OpenAI code migrates by changing only key, base URL and model name, and lists DeepSeek, Kimi, GLM and MiniMax as supported."},{"url":"https://docs.siliconflow.com/en/userguide/introduction","date":"undated","support":"SiliconFlow's documentation (no date) offers pay-as-you-go APIs for hosted open models such as DeepSeek-R1 and Qwen3-Coder."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.caixinglobal.com/2026-06-17/siliconflow-raises-294-million-as-chinas-ai-inference-demand-surges-102454833.html","date":"2026-06-17","support":"Caixin names ByteDance and Alibaba Cloud as ahead of SiliconFlow in 2025 (17 Jun 2026), so larger rivals can absorb its volume."}]}}}}},"novita-ai":{},"friendli-ai":{},"thinking-machines":{},"infinigence-ai":{"a":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"infinigence-ai","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"china-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://www.yicaiglobal.com/news/tencent-bytedance-other-chinese-tech-giants-launch-deepseek-models-via-cloud-platforms","date":"2025-02-05","support":"Yicai: Baidu, Alibaba, Tencent Cloud and Volcano Engine all launched DeepSeek-R1/V3 deployment within days of each other (Feb 2025)."},{"url":"https://www.news.cn/finance/20250213/1d369f19be5f42c29b150bc628f8ed05/c.html","date":"2025-02-13","support":"Xinhua/MIIT: all three basic telecom operators fully integrated DeepSeek open models; China Telecom and China Mobile on domestic Ascend stacks."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC: over the past two years price was almost always buyers' top selection factor for China MaaS; buyers usually call model APIs without secondary tuning."},{"url":"https://docs.infini-ai.com/","date":"undated","support":"Infinigence docs (undated): GenStudio MaaS uses OpenAI- and Anthropic-compatible interfaces."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC 2025 ranking by calls: Volcano Engine nearly half, then Alibaba Cloud, Baidu AI Cloud, SiliconFlow, China Mobile Cloud."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-inference-serving","entityId":"infinigence-ai","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"china-serving":{"alternatives":{"level":1,"evidence":[{"url":"https://www.idc.com/?p=114742","date":"2026-05-07","support":"IDC ranks China public-cloud MaaS by 2025 call volume: Volcano Engine, Alibaba Cloud, Baidu AI Cloud, SiliconFlow, China Mobile Cloud, and also names Tencent, SenseTime, Huawei and China Telecom (7 May 2026)."},{"url":"https://global.chinadaily.com.cn/a/202605/11/WS6a01b570a310d6866eb47fab.html","date":"2026-05-11","support":"China Daily cites national data that China's average daily token use passed 140 trillion by March 2026; Infinigence announced a 700 million yuan investment (7 May 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.alibabacloud.com/help/en/model-studio/compatibility-of-openai-with-dashscope","date":"2026-10-13","support":"Alibaba Cloud's documentation (updated Oct 13, 2026) says OpenAI code migrates by changing only key, base URL and model name, and lists DeepSeek, Kimi, GLM and MiniMax as supported."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.caixinglobal.com/2026-06-17/siliconflow-raises-294-million-as-chinas-ai-inference-demand-surges-102454833.html","date":"2026-06-17","support":"Caixin's page digest says SiliconFlow processes about 1 trillion tokens a day (17 Jun 2026)."},{"url":"https://eu.36kr.com/en/p/3901197105071748","date":"2026-07-18","support":"36Kr reports PPIO above 1.2 trillion tokens a day by June 2026 (18 Jul 2026): independents of that size can absorb a small player's volume."}]}}}}},"chutes":{}},"validation":[]}