Top economist warns AI profits depend on investor cash, not customer demand

Torsten Slok's analysis exposes a structural inversion in AI's economics. Traditionally, companies selling end products to consumers capture the highest profit margins; upstream suppliers earn less. But in AI, chipmakers command 41% margins while model and application makers operate at -59%—the losses are subsidized entirely by investor capital, not customer revenue. This creates a dependency loop: silicon manufacturers depend on continued capital raises from money-losing AI companies, not on organic end-user demand.

The Bank of International Settlements and other observers have flagged this dynamic as unsustainable. Hyperscalers are issuing record debt (Goldman Sachs projects AI investments beyond $1 trillion in 2026) while productivity gains remain narrow and concentrated in the Magnificent Seven. Tech writer Ed Zitron highlighted Oracle as a concrete case: negative $23.7 billion cash flow propped up by a $300 billion commitment to OpenAI, with $260 billion in lease obligations for infrastructure projects not yet underway. Should OpenAI, Microsoft, Google, Amazon, or Meta reduce their quarterly spend on GPUs and data center construction—a plausible scenario if customer ROI disappoints—the entire chain faces a shock. Slok's central question is whether AI's end customers will see returns fast enough to justify the spending; the risk is that they will not, and that investor patience will run out first.