Leaked documents shed light into how much OpenAI pays Microsoft

As the spotlight intensifies on OpenAI amidst a year filled with significant deal-making and speculation regarding a potential IPO, new insights have emerged into the company's financial dealings. Tech blogger Ed Zitron has disclosed leaked documents that shed light on OpenAI's revenue and computing expenses over the past years. According to the documents, in 2024, Microsoft received approximately $493.8 million from OpenAI as part of their revenue-sharing agreement. This figure saw a substantial increase in the first three quarters of 2025, reaching around $865.8 million. OpenAI reportedly allocates 20% of its revenue to Microsoft, stemming from a deal that involved an investment of over $13 billion from Microsoft into the AI startup. However, neither OpenAI nor Microsoft has publicly verified this revenue share percentage. The situation becomes more intricate as Microsoft also shares revenue with OpenAI, returning about 20% of the earnings generated from its Bing and Azure OpenAI Service. A source informed TechCrunch that the leaked figures pertain to Microsoft's net revenue share and do not account for the payments Microsoft makes to OpenAI from Bing and Azure royalties, complicating the overall revenue landscape. Financial reports from Microsoft do not specify the income derived from Bing and Azure OpenAI, making it challenging to assess the total financial interaction between the two entities. The leaked documents not only highlight the revenue figures but also provide a glimpse into OpenAI's expenditure relative to its income. Based on the 20% revenue-sharing figure, it can be estimated that OpenAI's revenue was at least $2.5 billion in 2024 and approximately $4.33 billion during the first three quarters of 2025, with indications that actual revenues may be higher. Further reports suggest OpenAI's revenue for 2024 could be around $4 billion, with $4.3 billion generated in the first half of 2025. CEO Sam Altman has indicated that OpenAI's revenue could surpass the previously reported $13 billion annually, potentially exceeding a $20 billion run rate by year-end and aiming for a staggering $100 billion by 2027. In terms of operational costs, Zitron estimated that OpenAI spent about $3.8 billion on inference in 2024, with this figure rising to roughly $8.65 billion in the first nine months of 2025. Inference refers to the computing power required to execute a trained AI model. Historically, OpenAI has depended heavily on Microsoft Azure for its computing needs, although partnerships with CoreWeave, Oracle, AWS, and Google Cloud have also been established. While OpenAI's training expenses are largely non-cash due to credits from Microsoft, the inference costs are predominantly cash-based. This raises questions about the sustainability of OpenAI's financial model, suggesting that it might be spending more on inference than it is earning. These revelations add fuel to ongoing discussions about the viability of investments in the booming AI sector, especially if a leading entity like OpenAI is operating at a loss. Both OpenAI and Microsoft have not provided comments regarding these disclosures.