OpenAI CEO Sam Altman has labeled the aggressive buildout of AI data centers "unsustainable silliness," taking aim at neocloud operators pouring billions into massive capacity without the enterprise demand or balance sheets to support it. While Altman insists OpenAI's own massive expansion remains profitable and backed by revenue, he contends that the broader infrastructure market is running on a reckless "cost is no object" mentality.
The real threat to infrastructure investors is the paradox of algorithmic efficiency. As frontier labs rapidly optimize model architectures and slash inference costs, today's multi-gigawatt facilities risk rapid obsolescence. If software improvements outpace compute demand, expensive multi-year power commitments and specialized hardware rapidly turn into illiquid, high-depreciation liabilities.
That dynamic exposes Big Tech and third-party neoclouds to severe counterparty risk. Long-term power purchase agreements and compute take-or-pay contracts could quickly turn toxic if end-user monetization cools, triggering a wave of debt defaults among overleveraged infrastructure operators. Although Altman downplayed OpenAI's vulnerability—and noted the lab will not resell its compute for now—his newfound caution marks a sharp pivot from the unconstrained capital spending that previously drew public barbs from Anthropic CEO Dario Amodei.