Torsten Slok, chief economist at Apollo Global Management, spent the past week issuing a string of warnings about the AI boom. His central claim: the numbers powering the AI trade are starting to wobble, just as AI-linked stocks have tumbled.

A Flat Curve Where a Steep One Was Expected

On Tuesday, Slok flagged a recent slowdown in the growth of token spending. In AI, tokens are the small chunks of text that models process, and customers typically pay by the token, which makes spending on them a direct read on usage. According to the research findings, this flattening had become visible by late October 2026, with token spending growth slowing over the period and becoming increasingly tied to pricing.

Slok described the demand curve for compute as flat and highly sensitive to price changes. A price-sensitive buyer, in his framing, is one who walks away the moment the bill goes up. If AI customers behave that way, providers cannot easily raise prices to recoup what they have spent.

Slok also appeared on Bloomberg on October 22, 2026, where he described a significant gap between the massive capital spending of hyperscalers — the largest cloud infrastructure operators — and the stagnant productivity seen across the broader economy.

The S&P 493 Problem

The sharpest part of Slok's argument concerns what he calls the S&P 493 — the S&P 500 minus the Magnificent Seven, the mega-cap tech names that have dominated returns. Under his analysis, those 493 companies have seen flat revenue per worker and flat profit margins despite their AI investments.

As a result, many firms are treating AI spending as a cost of doing business rather than a growth engine. The research summary also cites data from Ramp, the corporate card and spend management company, whose figures show AI-related spending increasingly concentrated among a small number of venture-backed companies. The research further notes that AI valuations face the risk of a “painful repricing” if expected returns fail to arrive quickly.

Capex at Macro Scale

Slok's projections put the scale of the bet in plain view: he expects hyperscaler capital expenditures could reach around 3% of US GDP annually between 2027 and 2029. The research findings note that this degree of investment echoes earlier economic booms that ended in market corrections.

Credit spreads — the extra yield investors demand to hold corporate debt over safer government benchmarks — are widening on AI-related debt, according to the research, and inflation is adding pressure as well. For data center projects that rely on borrowed money, that can mean delays. Slok's analysis warns that critical projects could be pushed back under these combined pressures, which would further cool enthusiasm for AI investment.

What It Means for Investors and the AI Trade

The most direct implication falls on stocks whose valuations rest on continued AI spending growth. If token demand really is flat and price-sensitive, revenue projections built on steadily climbing usage could look optimistic in hindsight.

The S&P 493 data points to a separate issue. Even if the hyperscalers keep spending, the ultimate justification for that spending is that other companies become more productive using AI. Flat revenue per worker and flat margins among those firms is the clearest sign that the promised productivity dividend has not arrived on schedule.

For traders, several signals are worth tracking closely: whether token spending growth reaccelerates or keeps slowing, whether credit spreads on AI-related debt continue to widen, and whether data center timelines start to slip.

Profit margins and revenue per worker across the S&P 493 may matter most of all. If those begin moving higher, the bulls get their evidence. If they stay flat while capex climbs toward 3% of GDP, the repricing risk Slok described becomes harder to dismiss.