Model

Microsoft books Azure revenue on Chinese open models running on Baseten hardware it does not own

Microsoft now bills customers' committed Azure dollars for open models from Chinese labs running on Baseten's hardware, a toll booth earning rent on compute it does not own.

Vincent JiangVincent Jiangtwo-minute read
Share
Satya Nadella, Microsoft's chief executive, in a black and white studio portrait
Satya Nadella, Microsoft's chief executive. His Azure catalog now bills Baseten-hosted Chinese open models as first-party Foundry consumption.

Microsoft's+0.23% — Microsoft, up 0.23 percent today model catalog now sells compute Microsoft does not own. On November 17, 2026, Baseten~$1.4B — Baseten, private, latest valuation ~$1.4B, a San Francisco inference company founded in 2019, valued at $13 billion after raising more than $2 billion, went live in public preview inside Microsoft Foundry, the Azure platform where enterprises build and govern AI applications 12. Four frontier open models, DeepSeek V4.1 Flash, GLM-5.3, GLM-5.3-Flash and Kimi K3, all run on Baseten's infrastructure, not Azure's own GPUs 13.

Every dollar of it counts against Azure commitments

The models are sold as first-party Foundry consumption, and 100% of the spend counts toward a customer's Microsoft Azure Consumption Commitment, the pre-purchased cloud dollars enterprises already owe 13. Microsoft books Azure revenue on hardware it did not buy, paid for out of contracts signed for its own capacity. Administrators opt in, deployments go live in seconds, requests are stateless with zero data retention, and general availability is promised for early 2027 12.

Microsoft's own capex has nearly tripled since mid-2024

  • Capex
  • Operating cash flow
$0B$20B$40B$60BQ3 '24Q1 '25Q3 '25Q1 '26Q3 '26$39.37B
Data
CapexOperating cash flow
Q2 '24$13.87B$37.2B
Q3 '24$14.92B$34.18B
Q4 '24$15.8B$22.29B
Q1 '25$16.75B$37.04B
Q2 '25$17.08B$42.65B
Q3 '25$19.39B$45.06B
Q4 '25$29.88B$35.76B
Q1 '26$30.88B$46.68B
Q2 '26$35.8B$55.44B
Q3 '26$39.37B$58B
Microsoft's capital spending and operating cash flow by quarter, in billions of dollars, from its SEC filings. Capex rose from $13.9 billion in the quarter ended June 2024 to $39.4 billion in the quarter ended September 2026, up 103% year over year, while operating cash flow rose from $37.2 billion to $58.0 billion. Source: Microsoft quarterly filings.9

Microsoft's own coding model wins on a price that expires in May

The launch lineup leans on Chinese labs: DeepSeek, Zhipu's GLM and Moonshot's Kimi. The same week, Microsoft shipped MAI Code 2, pitching 96% of Claude Opus 5.5's performance at 5% of the cost, but that math rests on a 50% promotional price that expires in May 2027 67. Independent of promotion, a community evaluation on Hacker News put DeepSeek V4.1 at 77.0 on DeepSWE 1.1 against 73.2 for MAI-2, a figure InfoWorld's report of Microsoft's own benchmark tables matches 57. The comparison is uncontrolled, as its author says.

A coding task on MAI Code 2 costs $0.31 only at the 50% promo price

  • Estimate
$0$2$4$6$8MAI-2 (promo)$0.309–0.618GPT-6.1 Sol$0.76Claude Opus 5.5$6.31
Data
ValueRange
MAI-2 (promo) (estimate)$0.31$0.309–0.618
GPT-6.1 Sol$0.76—
Claude Opus 5.5$6.31—
Average cost per coding task across four coding benchmarks, in dollars, as reported for Microsoft's MAI Code 2 launch. The MAI-2 bar is an estimate at 50% promotional pricing, which expires in May 2027; the whisker shows the roughly $0.62 a task the same comparison implies at list price. Sources: Brave New Coin, InfoWorld.6,7

Baseten gets the rails, Microsoft carries the question

Baseten gets Microsoft's distribution and billing rails, on top of annualized revenue the Wall Street Journal reported at around $600 million in the spring 8. The Chinese labs get first-party placement in the world's largest enterprise cloud. Enterprises get cheaper frontier intelligence on an invoice that credits their commitments. Microsoft shareholders hold the open question: whether a pass-through with no disclosed margin split earns more than renting Azure's own GPUs. Watch the first Azure bills under this arrangement, and whether MAI's discounted pricing holds after May 2027 17.

Get Model by email, free

The AI labs, their models, and the business behind them. In our daily email, every weekday.

Loading MSFT

Reader comments

0 comments

    Model

    HeyGen's Avatar 6 scored 84% on realism, but its 42.2% pick rate was below chance

    Avatar 6 launched November 17, 2026, with an 84% realism figure and first place in a blind arena; the score doubles a below-chance pick rate, and the arena capped 97% of clips at 35 seconds. That is the evidence base under a company claiming $200 million of annual revenue.

    Model

    MiniMax hit a record low, about 90% below its January peak, as a US rule gates its rented Nvidia compute

    Washington's new remote-access rule puts a US license between China's listed model makers and the offshore Nvidia capacity they rent. MiniMax, the purest listed bet on Chinese foundation models, holds the price tag.

    Model

    OpenAI cuts flagship pricing 20% to match Anthropic as heavy users call GPT-6.2 Astra unusable

    OpenAI cut its flagship's price to match Claude Fable 5.2's rate, and the users who run models hardest rejected it, 24 hours before Anthropic priced a $1.4 trillion IPO on the assumption the quality gap holds. The independent labs say the gap is real, for now.

    Model

    Alibaba stops giving away its best AI model, three days before earnings

    Alibaba launched Qwen4-Max on November 16 and kept the 2.4-trillion-parameter flagship behind its own paid API, promising open weights only for smaller models. Thursday's report is the first test of whether Qwen's three billion downloads become cloud revenue, and it lands two days after a US rule put the chips behind the model under license.

    Model

    Perplexity claims $1 billion of revenue on five times less spend than Anthropic's $1.4 trillion IPO

    Perplexity says it hit $1 billion in recurring revenue on a fraction of Anthropic's spend, three days before Anthropic prices a $1.4 trillion IPO on the opposite assumption. Whose economics hold will only be settled by a filing Srinivas now says could come in early 2027.

    Model

    Anthropic prices a $1.4 trillion IPO as Databricks logs show its costliest model at 0.6% of enterprise tokens

    Databricks' gateway logs show the costliest Claude tier at 0.6% of enterprise tokens and open-weight models near a third, hours before Anthropic prices a $1.4 trillion debut on premium-model revenue.

    Subscribe to TheInference for free