Leaked Documents Show OpenAI Inference Costs Exceeding Revenues

Leaked financial documents obtained by tech blogger Ed Zitron and reviewed by the Financial Times suggest that OpenAI is spending more to run its AI models than it earns in revenue. The data indicates that inference costs—the compute used to generate AI responses—climbed from $3.8 billion in 2024 to $8.65 billion in the first nine months of 2025. Specifically, the company reportedly spent $5.02 billion on inference with Microsoft Azure in the first half of 2025 alone.

The leaks also detail a mutual 20% revenue-sharing agreement between OpenAI and Microsoft. OpenAI paid Microsoft $493.8 million in revenue shares in 2024, a figure that rose to $865.8 million through the first three quarters of 2025. In return, Microsoft returns approximately 20% of revenues generated from Bing and the Azure OpenAI Service to the startup.

These findings challenge public statements made by CEO Sam Altman, who characterized the company's annual revenue as exceeding $13 billion, with a potential run rate over $20 billion by the end of 2025. The leaked data implies that actual revenues for 2024 and early 2025 were billions of dollars lower than reported. While a Microsoft spokeswoman stated the figures aren't quite right, the company declined to provide specific corrections.