OpenAI‘s Compute Costs surge, Raising Questions About profitability
San Francisco, CA – November 14, 2025 – Leaked documents reveal the escalating costs OpenAI is incurring to power its artificial intelligence models, especially for inference - the process of running trained models to generate responses. The data suggests OpenAI’s inference expenses may exceed its current revenue, fueling debate about the sustainability of valuations in the rapidly expanding AI sector.
openai currently boasts an annualized revenue run rate of $20 billion, with projections estimating potential revenue reaching $100 billion by 2027. However, the leaked figures paint a more complex financial picture. In 2024, OpenAI spent approximately $3.8 billion on inference,a figure that jumped to roughly $8.65 billion in the first nine months of 2025 alone.
Historically, Microsoft Azure has been OpenAI’s primary compute provider. The company has recently diversified,establishing deals with CoreWeave,Oracle,AWS,and Google Cloud to supplement its Azure reliance. Previous reports indicated a total compute spend of roughly $5.6 billion for 2024, with a “cost of revenue” reaching $2.5 billion in the first half of 2025.
A source familiar with the matter clarified that while OpenAI’s training costs are largely offset by credits from Microsoft as part of their investment, inference costs are predominantly cash expenses. Training refers to the initial computational resources needed to develop a model.
These numbers raise concerns that OpenAI may be spending more to operate its models than it currently earns in revenue,perhaps adding fuel to ongoing discussions about a possible “AI bubble.” The financial implications for other AI companies,manny of which have received substantial investments at high valuations,remain uncertain.
openai declined to comment on the leaked data. Microsoft did not respond to a request for comment from TechCrunch.
