After a year of rapid deal-making and buzz about a potential IPO, OpenAI is facing increased financial scrutiny. Tech blogger Ed Zitron has obtained leaked documents that shed light on OpenAI’s financials, particularly its revenue and compute costs over the past couple of years.
Zitron reported this week that Microsoft received $493.8 million in revenue share payments from OpenAI in 2024. This figure surged to $865.8 million in the first three quarters of 2025, according to the documents he reviewed.
OpenAI reportedly shares 20% of its revenue with Microsoft as part of a previous agreement in which the tech giant invested over $13 billion in the AI startup. (Neither OpenAI nor Microsoft has publicly confirmed this revenue-sharing percentage.)
However, the situation becomes a bit complicated because Microsoft also shares revenue with OpenAI, refunding about 20% of the revenues from Bing and Azure OpenAI Service, according to a source familiar with the matter who spoke to TechCrunch. Bing, which is powered by OpenAI, and the OpenAI Service provide developers and businesses with cloud access to OpenAI’s models.
The source further explained to TechCrunch that the leaked payments refer to Microsoft’s net revenue share, excluding what Microsoft pays OpenAI from Bing and Azure royalties. Microsoft accounts for those amounts when reporting its internal revenue share figures.
Microsoft doesn’t specify its earnings from Bing and Azure OpenAI in its financial reports, making it difficult to estimate how much the tech giant is returning to OpenAI.
Nonetheless, the leaked documents offer a glimpse into the hottest company in the private market today—not just in terms of revenue but also expenses.
Based on that widely referenced 20% revenue-sharing statistic, we can estimate that OpenAI’s revenue was at least $2.5 billion in 2024 and $4.33 billion in the first three quarters of 2025—though the actual figures could be higher. Previous reports from The Information suggested OpenAI’s 2024 revenue was around $4 billion, with about $4.3 billion generated in the first half of 2025.
Altman has also mentioned that OpenAI’s revenue is “well more” than reports of $13 billion annually and anticipates ending the year with an annualized revenue run rate exceeding $20 billion. He even suggested it could reach $100 billion by 2027.
According to Zitron’s analysis, OpenAI possibly spent around $3.8 billion on inference in 2024, increasing to approximately $8.65 billion in the first nine months of 2025. Inference refers to the compute used to run a trained AI model to generate responses.
Historically, OpenAI has primarily relied on Microsoft Azure for compute access, although it has also formed partnerships with CoreWeave, Oracle, and more recently AWS and Google Cloud.
Other reports have estimated OpenAI’s total compute expenses to be around $5.6 billion for 2024, with its “cost of revenue” at $2.5 billion for the first half of 2025.
A source shared with TechCrunch that while OpenAI’s training expenses are mostly non-cash—covered by credits from Microsoft—its inference spending is largely cash-based. (Training implies the compute resources required to initially train a model.)
Even though these figures don’t provide a complete view, they suggest that OpenAI might be spending more on inference than it earns in revenue.
These implications are likely to fuel ongoing discussions about the AI bubble, which has been a hot topic from New York City to Silicon Valley. If a major player like OpenAI is indeed operating at a loss while running its models, what does this mean for the significant investments and lofty valuations across the AI sector?
OpenAI has declined to comment, and Microsoft did not respond to TechCrunch’s request for remarks.
If you have any sensitive information or confidential documents, we’re investigating the inner workings of the AI industry—from influential companies to the individuals affected by their choices. Feel free to contact Rebecca Bellan at [email protected] or Russell Brandom at [email protected]. For secure communication, you can also reach them via Signal at @rebeccabellan.491 and russellbrandom.49.
