Torsten Slok, Chief Economist at Apollo Global Management, has issued a warning that the valuations of many AI companies are at risk of a "painful repricing" if the anticipated productivity gains from artificial intelligence do not translate into broader profit margin improvements as quickly as the market expects. Slok highlighted a significant divergence between aggressive, front-loaded valuations for AI companies and a much slower reality regarding cash flow generation outside the tech sector. He suggests that equity markets betting on instant earnings growth could face a repricing if it takes five years, rather than five months, for AI to significantly boost productivity.
Slok pointed out that there is currently no evidence of rising profit margins in the S&P 493 (the S&P 500 excluding the largest tech companies), which is crucial because AI company valuations implicitly assume that these margins will eventually climb. He argues that while technology companies can integrate AI into their products and processes almost immediately, most other sectors, especially capital-intensive and heavily regulated ones like healthcare, banking, manufacturing, and energy, face much longer implementation timelines due to regulatory hurdles, data governance requirements, and the need for extensive process re-engineering. This delay means that structural productivity gains could be pushed back significantly.
The economist also emphasized that companies will likely reduce their AI spending if they do not see a quick return on investment. He views the current industry focus on token optimization, model routing, and token marketplaces as an early indicator that AI implementation could be more challenging and slower than current market projections suggest. Slok cautioned that a mismatch between current earnings expectations and the actual time required for firms to generate ROI from AI investments could severely impact the valuations of many AI companies today. He added that falling token costs, while expanding usage, might also compress the revenue pool, questioning if all hyperscalers and model providers will achieve the returns implied by present valuations.
