Skip to main content

Microsoft pushes its own AI models at Ignite; OpenAI is 10.8% of its revenue

Microsoft launched Code 2 and Cyber 2 Flash at Ignite. Its filing shows OpenAI was $10.0 billion of quarterly revenue and owed it $7.2 billion.

Layloma Rashid
7 minute read

Review editor: Omer SheikhFact-checking editor: Naba Fatima

The Microsoft logo and name on the upper floors of a red-brick and glass office building against a blue sky

A Microsoft office in the Boston area. The company showed new AI models of its own at its Ignite conference. Photo: Praswin Prakashan / Unsplash.

Microsoft used its Ignite conference on Tuesday, November 17, 2026, to put its own AI models in front of customers. Chief executive Satya Nadella announced a new family of coding models and a security model, pointed to a chip Microsoft designed itself, and put models from OpenAI and Meta on the same platform.

The company's filings show why it wants more than one supplier. OpenAI accounted for $10.0 billion of Microsoft's revenue last quarter, which is 10.8% of the total, and it owed Microsoft $7.2 billion at the end of September, TECHi's calculation from the quarterly report shows. Microsoft shares were down 2.3% in afternoon trading, against a 1.2% fall in the Nasdaq 100 fund.

Article Brief

Key Takeaways

4 Points24s Read
  • The newsMicrosoft launched its own Code 2 coding models and a Cyber 2 Flash security model, and added OpenAI’s GPT 6.2 and Meta’s models to its Foundry platform.
  • The dependenceOpenAI accounted for $10.0 billion of Microsoft’s $92.3 billion of revenue last quarter, or 10.8%.
  • The receivableOpenAI owed Microsoft $7.2 billion at September 30, up 20% in three months and 11.6% of all receivables.
  • The billMicrosoft spent $39.4 billion on property and equipment in the quarter, about double a year earlier and 43% of revenue.

What did Microsoft announce?

The headline products are models that carry Microsoft's own name. Code 2 is a family of coding models that Nadella said sits at the frontier of cost and performance for everyday enterprise work. Code 2 Flash is the cheaper version for high-volume tasks. Both are available now in GitHub Copilot. Cyber 2 Flash is a security model, which Microsoft is offering in preview.

Nadella described Code 2 and Cyber 2 Flash as doing frontier-level work at a fraction of the cost. Microsoft did not publish prices in the keynote.

It also widened the menu of other companies' models. OpenAI's GPT 6.2 is coming to Foundry, Microsoft's platform for building AI applications, and Nadella said Meta's latest models are coming too. Microsoft's summary of the event says Foundry now offers more than 11,000 models. A smaller model for routing and classification, Microsoft-Decision-1.1, became generally available, and Nadella said more than 4,000 customers have used it.

Two other announcements point the same way. Microsoft will give trusted customers access to the weights of its models, meaning the underlying parameters, under a program it calls MAI Licensed Weights. And it has a new chip. Nadella said Maia 300 will deliver 40% more tokens per watt than leading systems built on graphics processors, which today mostly means Nvidia's.

Its Cobalt 200 virtual machines, built on Microsoft's own Arm-based processor, are now generally available and due in 24 Azure regions by the end of the year.

On the product side, Microsoft said it now has more than 50 million AI agents registered in Agent 365, its control system for agents, and announced a partnership to bring Salesforce into Copilot.

The rest of the keynote was about running and policing all of this. Perception, a security system built from teams of AI agents that attack, detect and repair, is now generally available. Agent 365 gains an analytics tool that Nadella compared to an accounting system for AI spending. And Microsoft launched an AI Transparency Initiative with 40 organisations, tied to standards work at the US National Institute of Standards and Technology.

How much of Microsoft's business is OpenAI?

More than most investors would guess from the headline numbers, and Microsoft now spells it out.

Because Microsoft holds about 24% of OpenAI, accounting rules treat OpenAI as a related party, and Microsoft's quarterly report discloses what the two do with each other. For the three months to September 30, Microsoft recorded $10.0 billion of revenue from its commercial arrangements with OpenAI, including revenue-sharing payments. Total revenue was $92.3 billion, up 18.8% from a year earlier, an increase of $14.6 billion. OpenAI was therefore 10.8% of the total, a pace of $40 billion a year from one customer.

The second number is what OpenAI owes. Accounts receivable from OpenAI were $7.2 billion at September 30, up from $6.0 billion three months earlier, a rise of 20%. That is 11.6% of everything Microsoft was owed by customers, against 7.4% in June. At the quarter's pace it equals 66 days of OpenAI revenue, a little more than the 62 days for Microsoft's customers as a whole. If OpenAI were paying at the average customer's pace, the balance would be about $6.7 billion.

Microsoft has committed $13.0 billion to its OpenAI investment and had funded $12.0 billion of that by the end of September. The investment added $207 million to net income in the quarter. A year earlier it had subtracted $3.1 billion.

Microsoft and OpenAI in the September quarter
MeasureFigureContext
Revenue from OpenAI$10.0B10.8% of total revenue of $92.3B
Receivable from OpenAI$7.2B11.6% of all receivables, up from 7.4% in June
Change in receivable since June+20%From $6.0B
Days of OpenAI revenue outstanding6662 for Microsoft’s customers as a whole
Stake in OpenAIAbout 24%$12.0B funded of a $13.0B commitment
Effect on net income+$207M−$3.1B a year earlier
Swipe across to compare all columns. Source: Microsoft Form 10-Q for the quarter ended September 30, 2026; TECHi calculations. Days outstanding are receivables divided by quarterly revenue, times 92 days.

Why does that make Ignite matter?

A customer that is one-tenth of revenue is a strength while it grows and a risk if it slows, switches suppliers or struggles to pay. OpenAI has been signing computing deals with other providers for more than a year, and it released a new model of its own on Tuesday.

Seen that way, the keynote reads as a list of hedges. In-house models for coding and security give Microsoft something to sell that does not depend on OpenAI's roadmap or share revenue with it. Hosting Meta's models next to OpenAI's makes Foundry a marketplace instead of a reseller. A chip of its own lowers the cost of running whichever model the customer picks. Even the small Decision model is built on Gemma, an open model from Google, according to Microsoft's event summary.

Microsoft's reporting has moved the same way. Since the start of this fiscal year it has reported two segments, one of which is called Agents and Infra.

The cost of the strategy is in the same filing. Microsoft spent $39.4 billion on property and equipment in the quarter, about double the $19.4 billion of a year earlier, and equal to 43% of revenue. It holds $683 billion of contracted future revenue, of which it expects to recognise about 30% within twelve months.

What does it mean for Microsoft stock?

Microsoft closed on Monday at $517.99. That values the company at about $3.85 trillion and is 26 times the $19.91 a share that analysts expect it to earn in the fiscal year to June 2027, according to estimates compiled by Nasdaq. The shares were up 7.1% for the year at Monday's close and 6.8% below their November 4 close of $555.80, the highest of the past year.

Tuesday's fall is hard to pin on the keynote. AI stocks were weak across the board ahead of Nvidia's results, though Microsoft fell roughly twice as much as the Nasdaq 100 fund.

The case for the strategy is simple. If customers buy Microsoft's cheaper models for routine work, Microsoft keeps more of each dollar and depends less on a partner that is also a competitor. The case against is that OpenAI's models are why many customers came to Azure in the first place. Selling against a company that supplies a tenth of your revenue is a delicate thing to do well.

TECHi noted in September that Microsoft had begun publishing Azure's revenue while keeping its margin private. The OpenAI figures are a second window that its filings have opened.

One quarter is also a thin base. The filing gives the OpenAI revenue figure for the latest quarter only, with no year-earlier number to set beside it.

What to watch

Prices for Code 2 are the first thing to look for, since a fraction of frontier cost can mean many things. The next quarterly report, due in late January, will show whether OpenAI's share of revenue and of receivables is still rising. Nvidia reports after Tuesday's close, and Microsoft is one of its largest customers as well as, with Maia, a would-be rival. TECHi's Microsoft quote page carries the live price.

Disclaimer

This article is for informational purposes only and does not constitute financial, investment, tax, or legal advice. Market data, tax rules, and prices can change after the article date. TECHi and its authors may hold positions in securities or digital assets mentioned. Always conduct your own research and consult a licensed financial, tax, or legal professional before making decisions.

Share

Pick your channel

Comments