Every enterprise AI pitch ends at the same place: higher revenue, fatter margins, more output per worker. Torsten Slok, partner and chief economist at Apollo Global Management, went looking for those three things in the aggregate data and published what he found on October 24. His Daily Spark note, "No Signs of AI Payoff Outside the Tech Sector", splits the S&P 500 into the Magnificent 7 and everyone else, the S&P 493, and tracks the two groups through the AI spending wave.
For the seven, the lines do what the pitch promises. Revenue per worker is approximately $780,000 for the Mag 7 and in the $525,000–$550,000 range for the S&P 493, according to secondary readings of Apollo's inflation-adjusted chart. Margins are approximately 30% and 10%, respectively, and the 493 lines have barely moved in about seven years. The San Francisco Fed productivity series cited by Slok shows no clear acceleration in aggregate total factor productivity growth; Slok's own reading is blunter, that there are still "no signs of AI boosting productivity growth". His summary is not hedged: the technology "has delivered zero financial payoff to corporate America outside of tech".
The buyers are a small room
The note's second finding explains part of the first, and it comes with a denominator warning. Ramp data cited by Slok, and described in secondary coverage of his charts, puts 99.5% of observed model-serving and inference spending among the top 10% of firms in the dataset. Of the top 1% of AI spenders in that dataset, 80.4% are reportedly venture-backed. Those percentages measure different things: the first is how concentrated the spending is, the second is who the heaviest spenders are. Neither is a census of corporate America. Together they say the heavy use of AI sits with companies that are either selling it or funded to experiment with it.
The mainstream buyer, on Slok's reading, is going the other way. "Mainstream corporate America is pulling back and starting to view token spending not as a source of growth but as a cost center that weighs on margins," he writes, adding that it "does not need the latest model and is unwilling to pay for it."
His explanation is qualitative, and worth quoting because it names the thing software demos leave out. "AI may be a great calculator, but in business, intelligence is mostly judgment," Slok writes. AI, he adds, "does not understand accountability, trust, culture, unwritten rules, ethical trade-offs or fairness, all of which shape reputational risk and ultimately drive revenue growth and profit margins."
What flat lines can and cannot show
Treat the word "zero" as an economist's headline, not a measurement. Part of the divergence may reflect AI expenditure flowing from customers to infrastructure and model suppliers. But the charts do not isolate those transactions, and much of the Mag 7's revenue and margin performance has other drivers. Margins also move for reasons that have nothing to do with AI, from layoffs to pricing power to the rate cycle. And an index average can sit still while real gains and real losses underneath it cancel out: a flat series can coexist with genuine cost avoidance, quality improvements and consumer surplus, including at firms outside the S&P 500 entirely.
Nor is this the first time Slok has run the test. In an August 9 Daily Spark note, "The Buyers of AI Are Still Waiting for the Payoff", he argued that the S&P 493 was spending heavily on AI without it showing up in profit margins. The October 24 note widens the evidence, adding revenue per worker, the productivity series and the spending concentration, and sharpens the language to "zero". The thesis is not new this weekend; the formulation is.
What survives those cautions is the standard of proof. The aggregates show no broad, identifiable acceleration in these measures, and these measures are the ones the enterprise AI pitch itself chose: margins and revenue per worker at the companies buying the tools. General-purpose technologies do pay late; electricity took decades to register in factory statistics. But "late" is a claim about the future, and it is currently carrying a great deal of present-tense market value. A stronger test would be repeatable, attributable results from identifiable non-tech adopters, ideally connected to financial or operational outcomes over time. Slok's closing line sets the venue: "The bottom line is that the AI story will be decided in the S&P 493, not the Mag 7."
Sources
- [1] Apollo Global Management, The Daily Spark (October 24, 2026) — Torsten Slok, "No Signs of AI Payoff Outside the Tech Sector": the zero-payoff claim, the token-spending, calculator-and-judgment and accountability passages and the closing S&P 493 line, all read first-hand by ALFW on Apollo's page; the note's charts are sourced to Bloomberg, Macrobond, the Federal Reserve Bank of San Francisco, the US Bureau of Labor Statistics and the Ramp AI IndexRead source
- [2] Apollo Global Management, The Daily Spark (August 9, 2026) — Torsten Slok, "The Buyers of AI Are Still Waiting for the Payoff": the earlier version of the same test, arguing S&P 493 AI spending was not showing up in profit margins, read first-hand by ALFWRead source
- [3] TechCrunch (October 25, 2026) — "AI delivers 'zero financial payoff' to most companies, Apollo economist says", coverage of the note with fresh Slok quotes; TechCrunch discloses in the article that Apollo is the private-equity firm that formerly owned TechCrunchRead source
- [4] PPC Land (October 25, 2026) — close reading of the note: data sources, the Ramp concentration figures and the closing lineRead source
- [5] Stocktwits / TradingView (October 25, 2026) — secondary readings of the note's charts: revenue per worker of roughly $780,000 against $525,000, profit margins of roughly 30% against 10%Read source
- [6] Brad DeLong, Substack (October 25, 2026) — "Torsten Slok Beats the AI-Bear Drum: Charts of the Day", quotations from the note including the token-spending passageRead source