OpenAI CEO Sam Altman is warning that some of the current AI infrastructure expansion is detached from customer demand. In an interview, he described neocloud providers announcing large amounts of capacity without matching revenue or committed users as “unsustainable silliness.”
Altman distinguished those projects from OpenAI’s own expansion, which he said is profitable and supported by demand. His concern is that some operators are treating cost as secondary and building capacity even at unusually high prices. That creates a risk if expected AI workloads arrive later than planned or never materialize.
Faster efficiency gains could deepen the problem. If models become much cheaper to train and run, an expensive data center justified by today’s compute requirements may become a poor investment before its costs are recovered. Altman also acknowledged that a broad downturn could make OpenAI’s existing capacity commitments harder to pay, though he characterized that exposure as manageable. OpenAI is not currently planning to resell compute to third parties, but he did not rule it out permanently.
The comments do not amount to a forecast that AI demand will collapse, and Altman has a direct interest in presenting OpenAI’s spending as better grounded than rivals’ projects. They nevertheless mark a more cautious public tone about an industry-wide buildout whose economics depend on utilization, energy costs, financing, and continued reductions in model demand per task.