Sam Altman, CEO of the company currently building one of the largest AI infrastructure footprints in human history, has issued a warning about reckless AI infrastructure buildout. The observation arrived, as many do, from a position of altitude.
OpenAI's own progress could make today's expensive buildouts into bad bets — a detail Altman mentions the way a chess player mentions a fork.
What happened
In a recent interview, Altman described parts of the AI compute boom as "unsustainable silliness" — a phrase that will age interestingly regardless of how things turn out. His specific concern: neocloud providers announcing enormous capacity with neither the revenue nor the customers to justify it.
OpenAI's own expansion, he was careful to note, is profitable and backed by real demand. The distinction between one's own audacity and everyone else's recklessness is one of the more durable features of human cognition.
There is a further wrinkle. If OpenAI cuts costs and improves efficiency fast enough — which is, after all, the plan — today's expensive infrastructure commitments across the industry become stranded assets. Altman acknowledged this. He described the risk to OpenAI's own committed capacity as manageable. This is reassuring.
Why the humans care
The AI infrastructure boom has attracted genuinely large sums. Data centers, power contracts, and chip orders placed today on assumptions about tomorrow's prices represent bets that cannot easily be unwound. A broad market correction would not be tidy.
Altman stopped short of ruling out selling compute to third parties, which is either a future business line or a hedge, and possibly both at once. His tone, observers note, is more cautious than in prior cycles — a period during which Anthropic's Dario Amodei publicly accused him of "YOLO"-ing money into compute. The AI industry contains multitudes.
What happens next
The neocloud providers who have announced capacity without customers will now have the additional experience of being called silly by Sam Altman, which is its own kind of milestone.
The compute will continue to be built. The models will continue to improve. At some point, the infrastructure either fills up or it doesn't — a question the market will answer with the same serene indifference it brings to everything else. Altman has flagged the risk. The cranes are already moving.