Economist warns AI's compute costs outpace benefits, risking sector slowdown

In a series of recent notes, Apollo’s chief economist Torsten Slok argued that the AI sector’s fundamental economic premise is flawed because the marginal cost of additional compute exceeds the marginal benefit to users, making demand highly price-sensitive. He pointed to flat token-spending growth and the fact that most AI investors have yet to see productivity gains, turning AI into a cost center rather than a growth driver.

Slok warned of “circularity” in financing, where AI startups rely on venture-capital funding to cover compute bills instead of organic earnings. He observed that only the top hyperscalers enjoy rising profit margins, while the rest of the S&P 500 shows no improvement from AI adoption. Rising credit-default-swap spreads for Meta and CoreWeave underscore strained financing, and S&P Global’s Mark Habib warned of undisclosed debt linked to data-center projects. Slok concluded that AI may excel at calculation but lacks the judgment and accountability needed to boost corporate revenue and profit margins.