OpenAI CEO Sam Altman is sounding a more cautious note on the AI infrastructure boom, according to The Decoder. In an interview cited by the outlet, Altman described parts of the global AI data center buildout as “unsustainable silliness,” pointing specifically to neocloud providers announcing large amounts of capacity without enough customers or revenue behind them. The core concern is not that AI compute demand is weak across the board. The Decoder reports that Altman said OpenAI’s own expansion is profitable and backed by real demand. His criticism was directed at companies pursuing a “cost is no object” approach to building capacity, including projects that assume very large compute demand will arrive at prices that may not hold. Altman also warned that OpenAI’s own progress could pressure the economics of today’s infrastructure bets. If compute costs fall and efficiency improves quickly, expensive projects financed under current cost assumptions could become bad investments, The Decoder reports. That is a direct risk for providers building capacity ahead of committed demand. The comments also cut back toward OpenAI’s own obligations. According to The Decoder, Altman acknowledged that a broad downturn could strain OpenAI’s ability to pay for capacity it has already committed to, though he considers that risk manageable. The summary does not provide figures for the commitments or name specific providers. Altman also said OpenAI is not selling compute to third parties yet, The Decoder reports, while not categorically ruling it out. That leaves open a potential future path in which OpenAI could treat some infrastructure as a product, but the provided reporting does not say the company has decided to do so. The caution is notable because it comes from one of the most aggressive buyers of AI infrastructure. The Decoder frames the comments as a shift in tone from earlier debates over compute spending, including Anthropic CEO Dario Amodei’s prior criticism that Altman was “YOLO”-ing money into the buildout. Based on the provided material, however, it is not clear whether Altman’s underlying strategy has changed or whether he is drawing a sharper distinction between OpenAI’s demand-backed spending and speculative capacity elsewhere. Who benefits: Providers with real customer demand and disciplined capacity planning are better positioned under Altman’s framing. AI customers could benefit if efficiency gains continue to push compute costs down. Who's exposed: Neocloud providers announcing large capacity without matching revenue or customers are the clearest exposure in the report. OpenAI also has some exposure through already committed capacity, though Altman reportedly called that risk manageable.