Microsoft Corp. quietly answered one of the most persistent questions hanging over the artificial intelligence boom last week, and the answer put a number on a worry that investors have been circling for months: much of the company's headline AI revenue is coming from a single customer it also helped create.
In a filing released the week of August 5, 2026, Microsoft disclosed that it recorded $24.1 billion in sales from OpenAI during the fiscal year that ended in June. Bloomberg, which first reported the figure, estimated that the sum represents more than half — and likely around 70% — of Microsoft's total AI business. It is the clearest look yet at how dependent the world's most valuable software company remains on the startup it has bankrolled since 2019.
What the filing actually shows
Microsoft has been unusually stingy with AI figures, disclosing its total AI business only twice: an annualized run rate of more than $13 billion for the quarter ending December 2024, and more than $37 billion for the quarter ending March 2026. To gauge OpenAI's slice, Bloomberg assumed Microsoft's AI run rate kept growing at the 123% clip reported for March, which would have produced roughly $34 billion of AI revenue for the full fiscal year. Against that, the newly disclosed $24.1 billion from OpenAI dominates the mix.
A company spokesperson confirmed the figure captures all sales and revenue share from OpenAI. Under their agreement, OpenAI pays Microsoft for cloud computing on Azure, for costs tied to building models, and hands over a share of its revenue. In other words, a large portion of the $24.1 billion is OpenAI's own compute bill — routed through Azure and booked as Microsoft income. The filing also showed Microsoft carried $6.0 billion in accounts receivable from OpenAI as of June 30, 2026.
Set against Microsoft's total company revenue of roughly $331.8 billion, OpenAI looks small — less than 10%. But as a driver of AI growth specifically, it is the engine.
The circularity problem
The disclosure matters because it sharpens a debate about how to read AI financials at all. When an investor and a customer are the same entity, revenue can start to look like a mirror. Microsoft has poured more than $13 billion into OpenAI; OpenAI turns around and spends heavily on Azure, which shows up as Microsoft AI revenue. Money invested returns as sales.
That pattern is not unique to Microsoft. In late July, Bloomberg reported that Nvidia's roughly $750 billion in interlocking commitments — including a proposed investment of as much as $100 billion in OpenAI — had revived fears of "AI circular financing." The mechanics rhyme: a chipmaker invests in a model lab, the lab signs cloud contracts, the cloud provider buys the chipmaker's GPUs, and cash that left as an "investment" returns as "revenue." Skeptics have compared it to dot-com-era round-tripping, in which companies inflated one another's growth. OpenAI's infrastructure commitments reportedly stretch toward $1.4 trillion even as it is on track to lose an estimated $14 billion in 2026.
Analysts say the key question is what kind of revenue Microsoft is actually booking. "One big remaining question is how much of OpenAI's contribution comes from the revenue-sharing agreement versus cloud computing or other Microsoft services," said KeyBanc analyst Jackson Ader. "The more of that revenue comes from services to OpenAI rather than the benefits of investment, the more favorably I'm going to look at it." The distinction is the whole ballgame: compute sold to a customer that would buy it anyway is durable demand; revenue share cycled back from an equity stake is closer to an accounting round trip.
Why disclose now, after years of silence? Olga Usvyatsky, an accounting researcher and founder of data-analytics firm Nonlinear Analytics, wrote in a note that the disclosure may be tied to OpenAI's plans for an eventual public listing — a step that would force far more of these interlocking flows into the open.
What to watch next
The $24.1 billion figure gives investors a concrete anchor for a business that had mostly been discussed in estimates, but it raises as many questions as it settles. Watch for three things. First, whether Microsoft breaks out how much of the OpenAI total is arm's-length cloud spend versus revenue share — the split Ader flagged. Second, any movement on OpenAI's IPO timeline, which would drag the broader web of Nvidia, Oracle and Microsoft commitments into regulatory daylight. And third, whether Microsoft's efforts to diversify — backing Anthropic and building its own models — begin to visibly dilute OpenAI's outsized share. Until then, the cleanest signal of real AI demand may be the revenue that comes from customers who are neither investors nor investees. For now, at Microsoft, that is the smaller number.
"The more of that revenue comes from services to OpenAI rather than the benefits of investment, the more favorably I'm going to look at it."- Jackson Ader, Analyst, KeyBanc