Sam Altman, CEO of OpenAI, stated at a press dinner in San Francisco on August 15, 2025, that the company would already be profitable from running AI models if they didn't have to pay for training new ones. Altman discussed in detail the period following GPT-5's release, as well as the company's future plans and financial outlook, emphasising that training costs for model development represent the greatest financial burden in the AI industry.
The CEO shared specific figures, stating that the company's logic is that by spending $300 billion they can sell $400 billion worth of services, adding that an increasing portion of computing capacity is now used not for training new models but for running existing ones (inference). Altman indicated that in the near future, OpenAI will likely spend trillions of dollars on building data centres, and that this massive investment will require innovative financing solutions. He declared that he suspect they can design a very interesting new kind of financial instrument for financing compute that the world has not yet figured out. Since GPT-5's launch, OpenAI's API traffic has doubled within 48 hours, indicating that demand remains extremely strong for the company's services.
OpenAI's enormous capital requirements will likely lead to a public stock offering in the future, although Altman noted he could imagine someone else leading the company during quarterly earnings announcements. The company's current business model can thus be divided into two parts: the already profitable operational business and the training area that continues to require significant investment, without which Altman claims OpenAI would be a very profitable company. This duality well illustrates the current challenges of the AI industry, where running models already generates revenue, while developing newer models continues to require enormous investments, significantly impacting companies' overall profitability.
Sources:
Maxwell Zeff
