Leaked Files Point to OpenAI’s Billion-Dollar Flow to Microsoft and a Murky 20 Percent Revenue Loop

Here’s the short version. Newly surfaced documents and media reports say OpenAI paid Microsoft hundreds of millions in 2024 and even more in early 2025, tied to a two-way revenue-sharing deal and massive cloud bills. Microsoft has pushed back on the exact numbers, but the leak spotlights a deeper truth. Keeping frontier AI online costs a fortune, and this partnership is the engine that makes it run.

A fresh batch of leaked financials, first highlighted by tech blogger Ed Zitron, claims OpenAI paid Microsoft about $493.8 million in 2024 and roughly $865.8 million in the first three quarters of 2025. Those payments map to compute and platform usage on Microsoft’s cloud, which powers products like ChatGPT for millions of users worldwide.

What is the 20 percent revenue-share everyone is talking about?

As TechCrunch reported, people familiar with the arrangement say OpenAI sends roughly 20 percent of its revenue to Microsoft. It is not just a one-way pipe though. Microsoft is also said to route around 20 percent of revenue from Bing Chat and the Azure OpenAI Service back to OpenAI. So, the money moves in both directions, and that makes the math a bit squishy when you try to figure out who is really ahead at any given moment.

Veteran Microsoft watchers, including Tom Warren, have noted that this reciprocal flow blurs the lines between partner and platform. Microsoft needs OpenAI’s most advanced models to differentiate Windows, Bing, GitHub, and Office. OpenAI needs Microsoft’s global cloud to serve those models at scale. It is a loop. Powerful, but complicated.

Editor’s pick

As Microsoft folds AI deeper into Windows and the cloud, having a PC ready for on-device AI features can help. The TechBull suggests looking at the Lenovo IdeaPad Slim 3X Copilot+ PC, built to tap Windows’ latest AI features.

Is OpenAI spending more to run AI than it brings in?

That is the worry. Zitron’s read of the leaked data suggests OpenAI’s inference costs reached about $8.65 billion in the first nine months of 2025. Inference is the meter that runs every time a user prompts a model. It is pricey because it leans on fleets of advanced GPUs and specialized infrastructure.

The cost problem is not unique to OpenAI. It is the structural challenge of the AI era. Training is expensive, yes, but serving billions of queries every day is the real budget-eater. The leaked figures simply put a sharper edge on what insiders already knew. A sustainable path likely requires cheaper inference, smarter caching, smaller tailored models for common tasks, and more efficient hardware. Big bets on next-gen data centers and custom silicon are already in motion, including mega-projects to scale AI infrastructure such as the rumored Stargate buildout referenced in our coverage of the OpenAI, Oracle, and SoftBank data center initiative.

Did Microsoft confirm the leaked numbers?

Not quite. Citing the Financial Times, The Register reported that Microsoft said the figures “aren’t quite right.” That careful phrasing keeps plenty of room for interpretation. The core dynamic still stands though. Microsoft is the indispensable cloud backbone for OpenAI, and OpenAI’s models power a wave of Microsoft products. The precise dollars and cents may be disputed, but the dependency is not.

The backdrop here is important. If OpenAI pursues an IPO, detailed financials would move from rumor to public record. These leaks serve as a preview of the scrutiny to come, and they raise fair questions investors will want answered with hard numbers.

What could this partnership mean for AI competition?

Deals at this scale tell a story. As Forbes’ Alex Konrad and others have argued, even the leaders in AI lean heavily on the big clouds to make their products real. Without Microsoft, Amazon, or Google, the cost and complexity of running frontier AI would be out of reach for most companies. That dependence raises the stakes for competition and for regulators who worry about choke points in compute, data, and distribution.

As The Times of India noted, the leaks underscore how intertwined these relationships have become. Expect more questions about market power, neutrality in cloud AI services, and how revenue-sharing might shape pricing and product choices across the ecosystem.

What should we watch next?

Three things. First, clarity. If OpenAI moves toward public markets, expect more concrete disclosures on revenue, unit economics, and the structure of payments with Microsoft. Second, costs. Watch for technical shifts that lower inference spend, from model optimization to new accelerators. Third, oversight. As the alliance deepens, antitrust and national security questions will likely intensify, especially around compute access and data flows.

For now, both companies are pressing ahead. OpenAI’s route to profitability runs through Microsoft’s data centers, and Microsoft’s AI ambitions hinge on OpenAI’s models. How they balance control, cost, and growth will set the tone for the next chapter of the AI industry.