Leaked Documents Suggest OpenAI May Still Be Spending More on Inference Than It Earns
OpenAI could still be losing money on inference despite rapid revenue growth, according to new internal financial documents shared publicly by tech blogger Ed Zitron. The leaked materials provide a rare look at the company’s compute expenses over the last two years and suggest that operational costs remain a major challenge.
What the Documents Reveal
The internal files outline OpenAI’s revenues from product subscriptions, enterprise contracts, and API usage, alongside its expenditures on GPU clusters, cloud resources, and data-center operations. Analysts reviewing the documents noted that inference — the computational process that powers ChatGPT and API responses — may be consuming more money than OpenAI is currently generating from those services.Inference Costs Outpacing Revenue Growth
While OpenAI’s earnings have grown significantly since 2023, the documents indicate that inference costs have grown just as quickly, if not faster. The expansion of ChatGPT usage, enterprise AI tools, and multimodal features requires immense GPU throughput, and scaling these workloads appears to be placing ongoing pressure on the company’s margins.Why the Gap Still Exists
Experts point out that OpenAI’s latest models require tens of thousands of high-end accelerators to operate at global scale. Despite partnerships with Microsoft Azure and ongoing efforts to optimize efficiency, the cost of inference for large multimodal models remains extraordinarily high. If usage spikes, expenses follow immediately.A Glimpse Into a Capital-Intensive Future
The leak underscores a growing concern within the AI industry: profitability for frontier-model companies remains uncertain. As OpenAI pursues even larger models and more complex product offerings, its reliance on massive compute infrastructure may continue to outpace its ability to monetize them in the short term.Editorial Team — CoinBotLab