Leaked docs show OpenAI paid Microsoft $866m in 2025 revenue share

Leaked documents show that OpenAI paid Microsoft US$493.8 million in revenue share for 2024, rising to US$865.8 million in the first three quarters of 2025, according to tech blogger Ed Zitron.

OpenAI reportedly shares about 20% of its revenue with Microsoft, though neither company has confirmed the percentage.

Sources told that Microsoft also shares revenue with OpenAI from Bing and Azure OpenAI Service, but specific amounts remain undisclosed.

Based on the reported 20% revenue-share, estimates suggest OpenAI’s revenue was at least US$2.5 billion in 2024 and US$4.3 billion in the first nine months of 2025, though some reports put the figures higher.

OpenAI’s spending on inference was estimated at US$3.8 billion in 2024 and US$8.7 billion in the first three quarters of 2025.

The company mainly relies on Microsoft Azure for compute but also works with CoreWeave, Oracle, AWS, and Google Cloud.

🔗 Source: TechCrunch

🧠 Food for thought

Implications, context, and why it matters.

OpenAI’s inference costs appear to exceed its revenues by a wide margin

OpenAI spent $8.67 billion on inference through September 2025, nearly double the at least $4.33 billion in revenue extrapolated from Microsoft’s revenue-share payments 1. The company appears to run losses on inference, which raises doubts about its model. (Inference refers to the compute used to run AI models in production; training is the compute used to build those models.)
Inference costs rose from $3.76 billion in 2024 to $8.67 billion by Q3 2025, which outpaced revenue growth in that span 1. CEO Sam Altman says revenue is “well more” than $13 billion, yet leaked files imply lower sums 1.
This gap between public revenue claims and inference spend puts OpenAI at the extreme end of cash intensive startups 1. Training bills are reportedly offset by Microsoft credits (free or discounted cloud compute), while inference outlays require cash, which strains finances as usage grows 2.

AI infrastructure operators can capitalize on demand for cost optimization tools

Rising inference bills spark demand for third-party AI FinOps (Financial Operations) platforms plus Large Language Model (LLM) gateways. These layers benchmark providers and route traffic to cheaper options. Startups that cut compute spend for enterprise AI teams can win customers looking to dodge OpenAI’s cost path.
OpenAI has long relied on Microsoft Azure for compute 2. It also has deals with CoreWeave and Oracle 2. The company works with AWS and Google Cloud, which marks a move toward multi-cloud (using more than one cloud provider) AI setups 2. Vendors and platform builders that simplify provider switching can win budget conscious buyers looking to trim inference costs.