Following a year filled with aggressive negotiation and speculation surrounding a potential IPO, the financial examination of OpenAI is becoming more rigorous. Confidential documents acquired by tech commentator Ed Zitron shed more light on OpenAI’s finances — particularly its earnings and computational expenses over recent years.  

Zitron disclosed this week that in 2024, Microsoft secured $493.8 million from OpenAI in revenue share payments. That figure soared to $865.8 million in the first three quarters of 2025, per the documents he reviewed.

OpenAI reportedly provides Microsoft with 20% of its revenue due to a prior agreement in which the software titan invested more than $13 billion in the influential AI startup. (Neither the startup nor individuals in Redmond have publicly validated this percentage.)

However, this situation gets somewhat complicated, as Microsoft also distributes revenue to OpenAI, contributing approximately 20% of the revenues from Bing and the Azure OpenAI Service, a source familiar with the situation informed TechCrunch. Bing operates with OpenAI’s technology, while the OpenAI Service markets cloud access to OpenAI’s models to developers and enterprises.  

The source also indicated to TechCrunch that the leaked payments pertain to Microsoft’s net revenue share, not the gross revenue share. In simpler terms, they don’t encompass what Microsoft paid to OpenAI through Bing and Azure OpenAI royalties. According to this source, Microsoft deducts those amounts from its internally reported revenue share figures.

Microsoft doesn’t specify how much it earns from Bing and Azure OpenAI in its financial reports, making it challenging to estimate how much the tech behemoth is returning.

Nonetheless, the leaked documents offer insights into the most sought-after company in the private market currently — revealing not only its revenue but also its expenditures in light of that revenue.  

Thus, using the widely circulated 20% revenue-share figure, it can be inferred that OpenAI’s revenue was at least $2.5 billion in 2024 and $4.33 billion during the first three quarters of 2025 — though it is likely higher. Previous reports from The Information estimated OpenAI’s 2024 revenue at about $4 billion, and its revenue in the first half of 2025 at $4.3 billion.  

Altman recently stated that OpenAI’s revenue exceeds reports of $13 billion annually, predicting it will surpass $20 billion in annualized revenue run rate by year-end (a projection, not guidance on actual revenue), with a possibility of reaching $100 billion by 2027. 

According to Zitron’s examination, OpenAI may have incurred approximately $3.8 billion in inference expenses in 2024. That expense grew to around $8.65 billion in the first nine months of 2025. Inference refers to the computing power utilized to operate a trained AI model for generating responses.  

Historically, OpenAI has predominantly depended on Microsoft Azure for compute access, although it has also established agreements with CoreWeave and Oracle, and more recently, with AWS and Google Cloud. 

Prior reports estimated OpenAI’s total compute expenditure at roughly $5.6 billion for 2024 and its “cost of revenue” at $2.5 billion for the initial half of 2025.  

A source knowledgeable about the matter told TechCrunch that while OpenAI’s training expenses are mostly non-cash — indicating they are covered by credits Microsoft granted OpenAI as part of its investment — the company’s inference expenses are largely cash-based. (Training pertains to the computing resources needed for the initial training of a model.)

Although this does not present a complete picture, these figures suggest that OpenAI might be spending more on inference costs than it brings in through revenue. 

These implications are likely to fuel the ongoing discussions about the AI bubble that have permeated conversations from New York City to Silicon Valley. If the model leader OpenAI continues to operate at a loss while running its models, what could this signify for the significant investments at astonishing valuations in the broader AI sector?

OpenAI declined to provide comments. Microsoft did not respond to TechCrunch’s inquiry for comment.

Do you have a confidential tip or documents? We’re reporting on the inner workings of the AI industry — from the companies shaping its future to those affected by their decisions. Reach out to Rebecca Bellan at [email protected] or Russell Brandom at [email protected]. For secure communication, you can contact them via Signal at @rebeccabellan.491 and russellbrandom.49.