The number is $24.1 billion — and most of it is OpenAI. Microsoft disclosed for the first time that OpenAI supplies roughly 70% of Microsoft’s AI revenue, according to filings reviewed by Rohan Paul. That figure includes the cloud bills OpenAI pays Microsoft for training and running ChatGPT in Microsoft data centers, plus model development costs and OpenAI’s own revenue share, all consolidated into Microsoft’s income statement.
The disclosure is a quiet landmark. Microsoft and OpenAI are the most scrutinized partnership in AI, and until now the financial dependency ran one direction in public: OpenAI needed Microsoft’s compute. The new filing shows the dependency runs both ways. Microsoft’s AI business is not a diversified portfolio; it is, to a large extent, a single customer.
What the $24.1 Billion Actually Includes
The 70% figure is not one line item. It bundles several revenue streams that all trace back to OpenAI:
- Azure cloud revenue from OpenAI training and inference workloads — the largest piece.
- Model development costs that Microsoft invoices as part of the partnership.
- OpenAI sales share, revenue from OpenAI’s own customers that flows through Microsoft’s books.
On the other side of the ledger, Microsoft has invested about $11.9 billion into OpenAI. So the relationship is structurally tight in both directions: Microsoft is OpenAI’s biggest landlord and investor, and OpenAI is Microsoft’s biggest AI tenant.
Why This Disclosure Matters
Three implications, in order of importance. First, Microsoft’s AI revenue growth is hostage to OpenAI’s spending, if OpenAI slows its training cadence or shifts compute, Microsoft’s headline AI number moves with it. Second, it puts the antitrust and regulatory scrutiny of the partnership in sharper focus: a 70% concentration would draw attention in any sector, and AI is the most scrutinized sector in tech right now. Third, it changes how investors should read Microsoft’s AI story, the “AI flywheel” pitch is really “the OpenAI flywheel.”
What to Watch
The interesting question is what Microsoft does about concentration risk. The company has been quietly building alternative capacity, its own frontier models and partnerships beyond OpenAI. The disclosure gives a baseline to measure that diversification against: if Microsoft’s non-OpenAI AI revenue grows faster than the total, the 70% will start to drift down, and that will be the signal that the hedge is working. Watch the Azure segment breakdown, not the total AI number, to spot the change early.