OpenAI’s finances have gone from expensive to eye-watering. The company reported $13.07 billion in revenue last year, but its net losses reached $38.53 billion, a scale that makes even fast-growing startups look comparatively restrained.
The numbers come from official financial filings cited by the Financial Times and Where’s You At, and they show how sharply OpenAI’s spending has accelerated. In the year before, OpenAI generated $3.7 billion in revenue and posted net losses of $5.09 billion. That jump tells the real story: revenue is climbing, but costs are running much faster.
OpenAI spending surge in 2025
OpenAI’s cost base ballooned across the board. Cost of revenue rose from $2.65 billion to $7.5 billion, research and development spending hit $19.18 billion, and marketing costs climbed more than fivefold to $5.73 billion. Total expenses reached $34 billion, while operating losses came to $20.92 billion.
There is a familiar pattern here: frontier AI companies often burn cash in pursuit of scale, but OpenAI’s pace is unusual even by Silicon Valley standards. If a startup is meant to be judged on momentum, this one certainly has it; if it is judged on profitability, the answer is less flattering.
Microsoft and SoftBank both benefited
The filings also show how much OpenAI’s ecosystem depends on its partners. SoftBank paid OpenAI $867 million for services in 2025, while Microsoft paid $303 million. At the same time, OpenAI paid Microsoft $10.59 billion for development and research services, likely tied to training AI models on Microsoft’s hardware infrastructure.
Microsoft ended the year having received $17.2 billion from OpenAI in total, with OpenAI still owing it $3.64 billion at year end. That is a tidy reminder that in the AI boom, the biggest near-term winners are often the infrastructure providers, not the model makers.
Why the OpenAI IPO conversation gets awkward
By the end of the year, OpenAI had $50 billion in assets, about half of them in cash, which gives it breathing room. But losses of $38.5 billion are not the sort of headline that helps a company pitch public-market discipline, especially when investors are already looking for proof that AI spending can eventually pay for itself.
The larger question is whether OpenAI is still in the phase where giant losses are forgiven as the price of building a category, or whether the market will start demanding something closer to a business model. If the filings are any guide, that argument is getting harder to avoid.