Microsoft's $37 Billion AI Business Has One Big Customer

Microsoft says its AI business has reached a $37 billion annual revenue run rate. That's up 123% from a year ago, and Satya Nadella calls it larger than some of Microsoft's biggest franchises. The number is impressive. It also has a problem.

OpenAI accounts for 45% of Microsoft's $625 billion revenue backlog as of December 2025 - $281 billion of contracted future spend, almost all of it Azure compute to train models and serve tokens. And in the 2025 calendar year alone, OpenAI paid Microsoft $17.2 billion for services: $10.6 billion in research and development costs (training), $6 billion in cost of revenue (inference), plus smaller charges for marketing and administration. OpenAI's revenue run rate is roughly $25 billion. It's spending more than two-thirds of its own income on Microsoft.

The way to understand Microsoft's AI growth is not to ask whether the number is large. It is. It is to ask how much of it is Microsoft selling AI to the world, and how much is Microsoft selling compute to the one company that already lives in its building.

Microsoft doesn't break out how much of its $37 billion AI run rate comes from OpenAI's own consumption versus third-party enterprise customers buying Azure OpenAI services, Copilot, or other AI products. But the arithmetic gives you a range. OpenAI spent roughly $13 billion on Azure compute in 2025, and Microsoft's AI run rate was $13 billion at the start of the year, growing to $37 billion by April 2026. If you assume OpenAI's Azure spend tracks roughly with its total Microsoft spend, the implication is that roughly 35% of Microsoft's AI revenue flows through a single customer.

This is not to say the revenue is fake. It's real money, real contracts, real infrastructure utilization. The question is whether a $37 billion AI business built around one hyperscaler customer is the same thing as a $37 billion AI business built on broad enterprise adoption.

Microsoft's broader AI adoption numbers are real but small. Microsoft 365 Copilot reached 30 million paid seats by the end of fiscal 2026. That sounds like a lot until you note that it still represents only a small fraction of Microsoft's commercial customer base. The product launched more than two years ago. Most of Microsoft's paying enterprise customers have not yet bought its AI add-on.

Meanwhile, Azure grew 43% in the latest quarter - a number that includes AI services, which contributed 12 to 16 percentage points of that growth, depending on the quarter. Without AI workloads, Azure would still be growing, but at a rate closer to 27-31%. The remaining growth comes from traditional cloud migration, database services, and enterprise software - all real demand, none of it particularly new.

There's another layer. Microsoft invested over $13 billion in OpenAI since 2019. OpenAI is now spending roughly the same amount back on Azure. The money Microsoft put in is, in effect, roundtripping through OpenAI and back into Microsoft's cloud revenue. Microsoft booked a $7.6 billion net gain from its OpenAI investment in Q2 FY2026 from investment markups. The same pool of money flows through both the investment gain and cloud revenue lines.

This isn't fraud. It's a circular relationship that inflates the appearance of an AI ecosystem. Microsoft is simultaneously the funder, the landlord, and the beneficiary. When one of your biggest customers is also your portfolio company, the line between organic growth and financial engineering gets blurry.

The market seems to understand this better than the headline numbers suggest. Microsoft's stock dropped 10% in January 2026 despite beating earnings, after investors fixated on the revenue backlog's OpenAI concentration and the $190 billion in planned capital expenditures. And Microsoft has been quietly diversifying: it's investing in Anthropic itself. Microsoft knows its bet on a single model provider carries risk.

The real test isn't whether the $37 billion number is big. It's whether it keeps growing if you remove OpenAI from the calculation. Microsoft's AI business is growing fast. But the faster question is whether the rest of it - the 30 million Copilot seats, the Azure OpenAI enterprise customers, the AI services that don't flow through OpenAI - is growing fast enough to justify a $190 billion infrastructure spend that Microsoft has to fund whether or not the demand materializes.

Here's how to think about it: if you strip OpenAI's Azure consumption from Microsoft's AI revenue, you're left with roughly $24 billion in third-party AI demand. That number is real, growing, and worth watching. It's also much smaller than the headline suggests. What matters is the direction: Microsoft's AI growth is heavily weighted to one relationship, and that concentration will either diversify or it won't. The next few earnings reports will tell you which.

The thing to watch isn't whether Azure grows 43% or 40%. It's whether Microsoft 365 Copilot's penetration moves from a small fraction to something above 10%. That's the signal that the rest of Microsoft's enterprise base actually wants AI. Until then, the $37 billion number is heavily dependent on one customer with a very large budget.