Is the AI bubble about to pop? Sam Altman is prepared either way.

Last Thursday, OpenAI CEO Sam Altman told reporters at a private dinner that investors are overexcited about AI models. "Someone" will lose a "phenomenal amount of money," he said, according to The Verge. The statement came as his company negotiates a secondary share sale at a $500 billion valuation—up from $300 billion just months earlier.

"Are we in a phase where investors as a whole are overexcited about AI? My opinion is yes," Altman told the journalists, comparing the current market to the dot-com crash of the 1990s. Wired reported that he also predicted his company will spend "trillions of dollars on data center construction in the not very distant future" and that ChatGPT will soon serve "billions of people a day."

For context, Facebook serves about 3 billion monthly active users. Altman's projection would require ChatGPT to reach nearly half the world's population as daily users (not monthly, like Facebook), which is an extraordinarily optimistic outlook.

Altman's bubble comments happened to land just before Fortune covered new MIT research showing widespread enterprise AI failures. The study, titled "GenAI Divide: State of AI in Business 2025," found that 95 percent of enterprise AI pilots fail to deliver rapid revenue acceleration. The research analyzed 300 public AI deployments, surveyed 350 employees, and included 150 interviews with business leaders, although Financial Times columnist Robert Armstrong noted this week that the MIT report "reads like something given away on the 'research' page of a large consultancy." Its conclusions are fairly obvious, he said: People like ChatGPT for basic tasks and hate complicated enterprise systems, and companies that try to build their own AI usually fail.

The study attributes these failures to implementation problems rather than model quality. "The core issue? Not the quality of the AI models, but the 'learning gap' for both tools and organizations," Fortune wrote about the study. Purchased AI tools succeed 67 percent of the time, while internally built systems succeed only one-third as often. This isn't necessarily an indictment of AI technology as a whole—it's potentially an indictment of corporate IT departments thinking they can out-engineer existing applications from AI service providers like OpenAI.