The $3 Trillion AI Bet and the Risks of a Hyperscale Hangover

Sequoia partner David Cahn, who first flagged the disconnect between AI hardware costs and revenue in 2023, now estimates that the sector must clear $3 trillion to recoup the costs of chips and data centers. This figure is likely conservative, as rising expenses for memory and specialized inference hardware continue to inflate the bill. While frontier labs report significant growth—OpenAI recently cited $20 billion in annualized run rate—the scaling pace of investment currently outstrips the velocity of product monetization.

Torsten Slok, chief economist at Apollo, warns that the market’s stability hinges on the assumption that hyperscalers like Google, Meta, Microsoft, and Amazon will see massive free-cash flow acceleration by 2028. However, a shift toward cheaper, open-weight models and increased token efficiency threatens those projections. If users continue to favor lower-cost alternatives without a compensatory explosion in total token consumption, the expected return on investment may fail to materialize. Slok cautions that if these cash flow targets are missed, the concentration of market value in a few tech giants could trigger a broader economic correction, moving the problem from a sector-specific concern to a systemic risk for the S&P 500.