Torsten Slok shows the AI buildout is a balance sheet fantasy
By Nikhil Raghavan · Reporting from San Francisco ·
Torsten Slok found that AI delivers no payoff to the S&P 493, proving that selling expensive tokens to non-tech firms is a sellable promise that fails on the balance sheet.
The current AI boom is a balance sheet fantasy. For years, the narrative has been that "intelligence" is a commodity that will automatically unlock productivity. But as Torsten Slok, chief economist at Apollo Global Management, recently pointed out, the math does not add up. The people selling the shovels are getting rich, but the people digging the holes are just paying a new, expensive tax. Jensen Huang claims demand is "off the charts," but he is confusing the desire for a miracle with a sustainable business model.
The token tax on corporate America
The gap between the hype and the ledger is stark. Slok analyzed the S&P 493, excluding major tech firms. He found that profit margins and revenue per worker for these companies are flat. According to reporting from TechCrunch and Yonhap, AI has delivered "zero financial payoff" to corporate America outside of tech.
The mechanism here is simple. Mainstream companies are spending on tokens, but they are not seeing a return. They are treating AI as a cost center that eats into their margins. Slok notes that while AI is a "great calculator," business requires judgment. AI cannot handle accountability, trust, or ethical trade-offs. Because of this, the promise of AI-driven productivity is sellable but not shippable. It looks good in a slide deck, but it fails when it hits the actual P&L.
A house of cards built on leases
This lack of productivity is not just a problem for non-tech companies. It is a systemic risk for infrastructure providers. CNBC reports that Meta entered negative free cash flow while holding over $450 billion in non-cancellable take-or-pay agreements. These are rigid, fixed costs that must be paid regardless of whether the AI actually works.
The "neoclouds" are in even worse shape. CoreWeave is a prime example of this fragility. Its credit default swap spreads have widened past 11%, while yields on its unsecured notes have topped 15%.
Financial filings show CoreWeave carries $35.6 billion in total debt obligations. The company also admitted that its internal controls over financial reporting were not effective as of December 31, 2025. When expected revenue fails to materialize, the credit analysts and CFOs of these neoclouds get paged at three in the morning.
The ghost of the dot-com bubble
We have seen this movie before. This is the Dot-com bubble all over again. The shared mechanism is a massive capital expenditure on infrastructure that precedes a productivity gap. Between 1995 and its peak, investments in the Nasdaq Composite stock market index rose by 600%. The index then fell 78% from its peak by October 2002. The market built the roads before it knew where the cars were going.
Today, we are replacing fiber optic cables with GPU clusters, but the result is the same. We are over-building for a demand that exists in the imagination of venture capitalists, not in the margins of the S&P 493.
The AI economy is currently a closed loop where tech giants sell hardware to neoclouds, who lease it back to enterprises that cannot find a way to make it profitable. This loop is held together by debt and the hope that "judgment" can be automated. It cannot. Until AI can actually drive revenue growth for a non-tech company, the current buildout is nothing more than an expensive exercise in over-leveraging.