Introduction
Leaked confidential documents are shedding new light on the complex financial relationship between OpenAI and Microsoft. Revelations published by tech blogger Ed Zitron show that in the first nine months of 2025, OpenAI paid Microsoft $865.8 million in revenue share payments, nearly double the $493.8 million for the entire year of 2024. These figures raise important questions about the economic sustainability of the business model of the most talked-about AI company of the moment.
The OpenAI-Microsoft Deal Structure
OpenAI shares 20% of its revenue with Microsoft as part of the agreement that saw the software giant invest over $13 billion in the startup. However, the financial dynamic is bidirectional: Microsoft kicks back approximately 20% of revenue generated from Bing (powered by OpenAI technology) and Azure OpenAI Service, a cloud platform offering access to OpenAI's models for developers and businesses.
A crucial aspect revealed by sources is that the leaked payments refer to Microsoft's net revenue share, not gross. In other words, the figures don't include what Microsoft has already paid to OpenAI from Bing and Azure OpenAI royalties. Microsoft internally deducts these amounts before calculating owed payments, making it difficult to estimate the exact money flow between the two entities.
OpenAI Revenue: Estimates and Projections
Based on the widely reported 20% revenue share percentage, it's possible to infer that OpenAI's revenue in 2024 was at least $2.5 billion, rising to $4.33 billion in the first three quarters of 2025. However, these figures likely represent a conservative estimate. Previous reports from The Information indicated 2024 revenue around $4 billion and first-half 2025 revenue at $4.3 billion.
Sam Altman, OpenAI's CEO, recently stated that the company's revenue is "well more" than the reported $13 billion annually, with projections indicating an annualized revenue run rate exceeding $20 billion by year-end. Ambitions are even higher: Altman predicts OpenAI could reach $100 billion in revenue by 2027.
The Inference Cost Problem
Inference costs represent the computational expense required to run an already-trained AI model and generate responses. According to Zitron's analysis, OpenAI spent approximately $3.8 billion on inference in 2024, a figure that skyrocketed to roughly $8.65 billion in the first nine months of 2025.
OpenAI has historically relied almost exclusively on Microsoft Azure for access to computational resources, though it has recently signed agreements with CoreWeave, Oracle, AWS, and Google Cloud as well. A source familiar with the matter told TechCrunch that while model training spend is primarily non-cash (paid through Microsoft credits allocated as part of the investment), inference spend is largely cash, meaning actual money paid out.
Previous reports estimated OpenAI's total compute spend at around $5.6 billion for 2024, with a "cost of revenue" of $2.5 billion in the first half of 2025. The new data suggests that inference costs alone could exceed the company's total revenue, raising serious questions about current profitability.
Implications for the AI Bubble
If OpenAI, the undisputed leader in the large language model sector, is still operating at a loss despite multi-billion-dollar revenue, what does this mean for the entire AI ecosystem? Can the massive investments at stratospheric valuations that characterize the sector be sustainable in the long term?
These questions fuel the increasingly heated debate about a possible AI bubble, a discussion that permeates every conversation from New York's financial world to Silicon Valley. The absence of official comments from OpenAI and Microsoft leaves room for interpretations and concerns about the economic model underlying the generative AI revolution.
Conclusion
The leaked documents offer a valuable, albeit partial, window into OpenAI's financial situation, the most watched private company of the moment. While revenue is growing rapidly, operating costs appear to be growing at an even faster pace, raising questions about the path to profitability. With the prospect of an IPO on the horizon and increasingly close scrutiny from investors and analysts, OpenAI will need to demonstrate that its business model can generate sustainable economic value, not just technological innovation.
FAQ
How much did OpenAI pay Microsoft in 2025?
OpenAI paid Microsoft $865.8 million in the first nine months of 2025 as revenue share, nearly double the $493.8 million for the entire year of 2024.
What is the revenue share percentage between OpenAI and Microsoft?
OpenAI shares 20% of its revenue with Microsoft, while Microsoft kicks back approximately 20% of Bing and Azure OpenAI Service revenue to OpenAI, creating a bidirectional financial flow.
What are OpenAI's estimated revenues?
Based on leaked payments, OpenAI's revenue was at least $2.5 billion in 2024 and $4.33 billion in the first three quarters of 2025, with projections exceeding $20 billion annualized.
What are inference costs for OpenAI?
Inference costs represent computational expenses to run AI models and generate responses. For OpenAI, these costs rose from $3.8 billion in 2024 to approximately $8.65 billion in the first nine months of 2025.
Is OpenAI losing money despite high revenue?
Leaked documents suggest that inference costs alone could exceed OpenAI's total revenue, indicating potential operating losses despite rapid revenue growth.
Which cloud providers does OpenAI use besides Microsoft?
In addition to Microsoft Azure, OpenAI has signed agreements with CoreWeave, Oracle, AWS, and Google Cloud to diversify access to computational resources.
Is OpenAI's training spend cash or non-cash?
Model training spend is primarily non-cash, paid through Microsoft credits, while inference spend is largely actual cash expenditure.
What do these figures imply for the AI bubble?
If OpenAI operates at a loss despite being the sector leader, doubts emerge about the sustainability of massive investments and high valuations across the entire AI ecosystem.