Here's the 'core economic flaw' in the AI bull case as stock and bond trades unravel, according to a top economist

Apollo chief economist Torsten Slok spent the week taking aim at the AI trade while Wall Street was already backing away from the sector. His main point was simple and uncomfortable: demand for compute doesn’t look limitless, and it seems highly sensitive to price. He said token spending growth has slowed, while many companies putting money into AI still haven’t seen meaningful productivity gains. To him, that makes AI look less like a growth engine and more like a cost line.

Slok’s sharper claim is that the economics don’t yet work. The marginal cost of producing compute, he argued, is running ahead of the benefit users actually get from it. That puts a hard cap on how much data-center capacity firms will want to buy, no matter how much optimism surrounds the buildout. In his view, the real driver is not physical capacity or power availability, but the price customers are willing to pay — and that price only makes sense if AI clearly lifts revenue, profits, or savings at scale. He said there is little sign of that so far.

Then came the issue of circularity. On Thursday, Slok warned about token spending that gets recycled through the same ecosystem, pointing to financing deals that have drawn scrutiny elsewhere in the AI world. He said a similar pattern shows up between AI startups and the venture capital firms backing them, noting that 80% of the top 1% of AI spenders are VC-backed. His read is blunt: startups are funding their compute bills with venture money, not with earnings from customers.

By Saturday, he was arguing that AI has delivered zero financial payoff to U.S. companies outside the big hyperscalers. The Magnificent Seven have seen profit margins and revenue per worker improve, he said, but the rest of the S&P 500 has not. That matters because Corporate America, in his telling, does not need the newest model and does not want to pay for it. And even if AI gets cheaper and more common, Slok doubts it will translate neatly into better business results, since judgment, trust, culture, and accountability still sit outside the machine.

All this landed just as AI stocks, especially chip names, sold off and traders started to question how long the capex binge can last. A handful of hyperscalers plan to spend an estimated $800 billion this year, with next year seen topping $1 trillion. But the financing side is getting tense too. Meta has issued a lot of debt, announced a new class of shares this week, and saw the cost of insuring its debt against default hit a record on Thursday. Smaller names are under pressure as well, with CoreWeave down more than 30% this month and its five-year credit default swaps jumping to 1,145 basis points. S&P’s Mark Habib also warned that there may be more debt tied up in these interconnected deals than investors can actually see.