Newly disclosed internal documents show that OpenAI paid Microsoft approximately $493.8 million in 2024 and about $865.8 million in the first nine months of 2025 under a reported 20% revenue-sharing agreement. Source documents suggest Microsoft also returns around 20% of its revenue from Bing and the Azure OpenAI Service back to OpenAI, but those payments are deducted internally before Microsoft reports net revenue-share figures, making the full financial flows opaque. The documents also reveal OpenAI’s compute spending is enormous—with inference costs of around $3.8 billion in 2024 and approximately $8.65 billion in the first nine months of 2025—raising questions about profitability despite high revenue growth. Meanwhile, separate reports indicate OpenAI plans to reduce the percentage of revenue shared with commercial partners (including Microsoft) to about 8% by decade’s end, potentially allowing it to retain over $50 billion more.
Sources: Economic Times, TechCrunch
Key Takeaways
– OpenAI gave Microsoft nearly half-a-billion dollars in 2024 and close to nine-hundred million in the first nine months of 2025 under a revenue-share clause reportedly at ~20%.
– The partnership remains highly asymmetric and opaque: Microsoft both receives payments from OpenAI and pays OpenAI via its Bing/Azure AI services, but internal deductions render net figures murky.
– OpenAI’s compute (inference) costs are ballooning—spending exceeded many of the payments to Microsoft—while the company anticipates reducing its payout percent to commercial partners down to ~8% by the end of the decade.
In-Depth
The recent leak of internal financial documents attributed to tech observer Ed Zitron has laid bare just how much the relationship between OpenAI and Microsoft has cost the former—and how much it continues to pay. According to the leaked files, OpenAI handed over roughly $493.8 million in 2024 to Microsoft and about $865.8 million during the first three quarters of 2025. These payments stem from a long-standing revenue-share agreement tied to a multibillion-dollar investment by Microsoft in OpenAI several years ago. Under that deal, OpenAI reportedly committed to transferring about 20% of its revenue to Microsoft. While neither company has publicly confirmed the exact figures, the documents give an unusually detailed look at the finances behind one of the most high-profile partnerships in the AI industry.
But the story grows more complex. Microsoft, in turn, reportedly pays OpenAI a slice of its revenue derived from the Bing search engine and the Azure OpenAI Service—which incorporates OpenAI’s models on Microsoft’s infrastructure. Sources in the TechCrunch report suggest Microsoft deducts those payments internally before reporting net revenue-share figures, meaning OpenAI may actually receive more than publicly observed—but tracking the flow becomes nearly impossible from outside the companies. The opacity raises questions about how much value OpenAI is really retaining and how much of its operational burden is subsidized via its relationship with Microsoft.
Meanwhile, OpenAI’s compute costs are staggering. The documents indicate inference costs alone—i.e., the process of running trained models to respond to queries—rose from about $3.8 billion in 2024 to nearly $8.65 billion in the first nine months of 2025. Considering the revenue payments to Microsoft and the high compute spend, the data suggest OpenAI may be burning through cash even as revenue grows. If a company is paying hundreds of millions to its partner, while infrastructure costs amount to multiple billions, questions of sustainable profitability naturally arise.
Adding another layer, earlier reporting indicates OpenAI intends to renegotiate its revenue-sharing terms. A Reuters article, citing The Information, reported OpenAI is aiming to reduce its payout to commercial partners—including Microsoft—from around 20% today down to approximately 8% by the end of the decade. That shift could allow OpenAI to retain tens of billions of dollars more—but may also test the tolerance of Microsoft, which continues to hold a significant stake and strategic tie to OpenAI. In October 2025, a restructured agreement between the pair saw Microsoft hold about 27% ownership in the newly formed for-profit OpenAI entity, while continuing to receive revenue-share payments of roughly 20% until specific milestones (e.g., achievement of artificial general intelligence) are met.
From a conservative perspective, the partnership underscores how deep and entangled large-scale tech-ecosystem deals have become—and how they hide both opportunity and risk. Microsoft essentially serves as the infrastructure backbone and early funder of OpenAI while also being a commercial customer. But the cost of scale is immense. OpenAI’s ambition to build AI at global scale requires massive capital, and its current cash flows suggest a reliance on external funding or favorable terms to stay afloat. Meanwhile, any renegotiation of sitting deals (like reducing payout percentages) must be handled carefully lest partner relations sour.
For investors, policymakers and taxpayers alike, the leaked numbers raise critical questions: At what point does AI investment cross into unsustainable territory? When a private company passes a large chunk of revenue to a partner while burning billions on compute, how much of the value generated accrues to end-users versus internal stakeholders? From a regulatory viewpoint, the depth of integration between Microsoft and OpenAI invites scrutiny about competition, control of compute capacity and future profits.
Ultimately, the leaked documents don’t just tell us how much OpenAI paid Microsoft—they tell us how much is at stake. The numbers underpinning the AI race are no longer abstract. OpenAI may be sprinting toward a future of transformative tech, but it is doing so at what may be a historically high cost—and the structure of its deal with Microsoft means both risk and reward are shared, but not equally. In that sense, the entire industry may be watching not only for the next model release, but for whether these financial engines hold up under pressure.