Most of Microsoft's AI revenue is just OpenAI paying Microsoft
A new Microsoft filing shows 24.1 billion dollars of revenue from OpenAI in the fiscal year ended June, which analysts read as roughly 70 percent of Microsoft's AI business. Outstanding receivables from OpenAI stood at 6 billion dollars.
Microsoft disclosed in a filing this week that it recorded 24.1 billion dollars in revenue from OpenAI during the fiscal year that ended in June. Bloomberg’s reading of the numbers is that OpenAI accounts for more than half, and possibly around 70 percent, of Microsoft’s entire AI business. The filing also shows 6 billion dollars in accounts receivable from OpenAI as of 30 June, meaning money billed but not yet paid.
The 24.1 billion figure is not one thing. It bundles payments for cloud computing capacity, contributions to model development costs, and Microsoft’s share of OpenAI’s revenue under their partnership agreement. It arrives while Microsoft is visibly trying to reduce its dependence on that one relationship: it has invested in Anthropic, and it has been shipping its own MAI models.
What is actually going on here
There is a pattern in this AI cycle that is worth naming, because it explains a lot of the very large numbers you see in headlines. Money moves in a loop. An AI lab raises capital, spends most of it renting compute from a cloud provider, and that spending shows up as the cloud provider’s AI revenue, which supports the cloud provider’s valuation, which helps fund further investment in the lab. None of this is improper, and the compute is real and genuinely expensive. But “AI revenue” as a line item means something different when a single customer is most of it.
For anyone trying to judge how big AI actually is as a business, the useful question is how much revenue comes from ordinary customers buying ordinary products, as opposed to labs paying each other. This filing is one of the first times a company that size has given enough detail to do that arithmetic from the outside, which is why analysts pounced on it. The 6 billion in receivables is the other detail people noticed, since it means a meaningful chunk of that revenue is recognised but not yet collected.
The counterpoint is straightforward: OpenAI is a very large customer with very large compute needs, concentration in a young market is normal, and Microsoft’s non-OpenAI AI revenue is still several billion dollars in absolute terms. This is a warning light, not a fire.
What this means for you: for your day to day use of Copilot or ChatGPT, nothing at all. The value here is as a reading aid. When you next see a company announce a huge AI revenue number, it is fair to ask who is paying it, and whether that customer is itself funded by the announcing company or its investors. That one question cuts through a surprising amount of the noise. If you make purchasing decisions, note also that Microsoft is hedging its own bets across OpenAI, Anthropic and in-house models, which is a reasonable posture for you to copy.
Sources
An open model is now months, not years, behind the frontier. Its safety testing is not.
A SaferAI report finds Z.ai's open-weight GLM-5.2 sits only a few months behind the leading closed models on cyber and biology capabilities, and refused none of the offensive tasks it was given. Claude Opus 4.7 refused so consistently the test could not be completed.
