Microsoft's $24.1B AI revenue mostly OpenAI's compute bill

Microsoft's shares climbed after earnings last week on the back of strong AI business growth, but a regulatory filing released alongside those results reveals an uncomfortable concentration: $24.1 billion of the company's AI revenue during the year to June came from OpenAI. By Bloomberg's estimate, that represents roughly 70% of Microsoft's actual AI sales—the clearest picture yet and the first time Microsoft has disclosed the figure explicitly, after years of investor pressure.

However, the revenue figure masks the real nature of the relationship. Under their agreement, much of the $24.1 billion is not new AI sales but rather OpenAI's own compute bills and model-development costs, which OpenAI pays to Microsoft and is routed through Azure and booked as Microsoft income. In other words, the growth in this revenue stream is partly a reflection of OpenAI's own spending rather than independent Microsoft business. Set against Microsoft's total company revenue, OpenAI shrinks to under 10% or closer to 7%, but when narrowed to the AI unit alone, the concentration jumps sharply into view.

The dependence extends into Microsoft's own operations. Jay Parikh, who runs Microsoft's CoreAI engineering group, sent a memo to staff instructing them to default to OpenAI's GPT-5.6 Sol model in GitHub Copilot for most use cases. His explicit reasoning: "shifting more workloads to OpenAI models helps us get greater value from our token investment." Microsoft holds intellectual-property rights to OpenAI's models through 2032, and by pointing its own engineers toward OpenAI's solutions, the company aims to extract more value from that stake. The memo also reflects a broader industry shift away from the brief era of "tokenmaxxing," when developers ran up large AI bills without monitoring output, as cheaper open-weight models—many from China—begin to pull spending in other directions.

Microsoft has not been passive in attempts to reduce this reliance. The company has built its own MAI models and offers cloud customers access to more than 11,000 models. It has also placed a $5 billion bet on Anthropic, OpenAI's largest competitor. GitHub Copilot, now past 50 million users, integrates models from Anthropic, Google, Moonshot, xAI, and Microsoft itself. Yet despite these efforts and over $260 billion in capital spending since 2022, roughly 70% of AI revenue still traces to OpenAI—a metric that has not budged. Analysts remain divided on whether this represents strength or fragility. KeyBanc's Jackson Ader wants clarity on whether the revenue stems from services sold to OpenAI or the benefit of Microsoft's own investment. Skeptics like writer Ed Zitron argue that much of the AI revenue is OpenAI's compute bill dressed as growth, and that if OpenAI stumbles, Microsoft will be left with sunk costs and empty data centers. For now, there is no sign of a stumble—the AI unit continues growing fast, which is why the stock climbed—but the filing settles a long argument with an uncomfortable answer: after years of work and hundreds of billions in spending, Microsoft remains, for now, mostly OpenAI's landlord.