Economy / Technology — Torsten Slok, partner and chief economist at Apollo Global Management, has put the artificial intelligence boom to its simplest possible test — and, on his numbers, corporate America outside the technology giants is failing it. Revenue per worker and profit margins are climbing for the Magnificent Seven. For the other 493 companies in the S&P 500, both lines are essentially flat.
Track it live: Live markets tracker
The Apollo Slok zero AI payoff argument landed on October 24, 2026, in a Daily Spark note titled “No Signs of AI Payoff Outside the Tech Sector,” and it was deliberately blunt. As Bloomberg reported the same day, Slok’s conclusion is that artificial intelligence has delivered “zero payoff” — in his fuller phrase, zero financial payoff — to corporate America outside the technology sector. It is an extraordinary claim about the most expensive technology build-out in corporate history. It is also, Slok insists, simply what the data show when you stop letting seven enormous companies speak for five hundred.
The test: seven companies against 493
Slok’s method is disarmingly simple. He splits the S&P 500 into two groups: the Magnificent Seven — the giant platform companies that design the chips, run the data centres and sell the models — and everybody else, the S&P 493. He then asks the two questions any genuine productivity revolution should eventually answer. Is each worker producing more revenue than before? And are companies keeping more of each dollar of sales as profit? For the Magnificent Seven, according to the charts in the October 24 note, drawn from Bloomberg and Macrobond data, the answer to both questions is yes: revenue per worker is rising and margins are expanding. For the 493, both measures are essentially flat. The note follows related Apollo analysis published on June 30 and August 9, 2026; October’s is the moment he stated the conclusion without hedging.
The gap between the two groups is not subtle. On Apollo’s figures, inflation-adjusted revenue per employee runs at roughly $780,000 for the Magnificent Seven against about $525,000 for the S&P 493 — a difference of nearly half again as much output value per worker. Profit margins are starker still: nearly 30 per cent for the Seven, against around 10 per cent for the rest of the index. The Seven were more productive and more profitable before the AI era; what Slok’s charts add is the direction of travel since the spending surge began. The giants are pulling further away. The broad corporate economy, where most Americans work, is not moving at all.
The productivity record agrees — so far
If AI were transforming ordinary businesses, the place it should appear first is the national productivity data, and Slok checked there too. The San Francisco Federal Reserve’s utilisation-adjusted total factor productivity series — the measure economists trust most for separating genuine efficiency gains from the noise of the business cycle — shows no clear AI-driven acceleration. Slok’s productivity charts draw on Federal Reserve and Bureau of Labor Statistics sources, and they tell the same story as his company-level work: no break in the trend, no kink in the line where the models arrived. Computers took years to surface in these statistics — the famous productivity paradox — but they did surface. On the evidence through October 2026, AI has not yet, outside the firms selling it.
Who is actually spending on AI
The spending data Apollo cites, from Ramp, may be the most revealing part of the whole argument, because it explains why the payoff is so concentrated. The top 10 per cent of firms account for 99.5 per cent of all spending on model serving and inference — meaning that the overwhelming majority of companies are barely spending anything on actually running AI in production at all. Drill into the heaviest users and the picture tilts further from the corporate mainstream: 80.4 per cent of the top 1 per cent of AI spenders are venture-backed companies, young firms burning investor capital to build AI-native products, not established businesses retooling how they work. Slok’s reading is that mainstream companies increasingly treat token spending — the per-query cost of using a model — as a cost centre to be minimised. They do not need the latest, most expensive model for the tasks they have found for AI, and they are unwilling to pay frontier prices for marginal improvements. Rational firm by firm, that behaviour produces exactly the flat lines Slok is charting.
An independent line of evidence points the same way. Forrester, as reported in analysis published on October 25, projects that AI will account for about a fifth of corporate technology spending in 2027 — a huge share of budgets — yet expects more than half of IT leaders to miss their AI return-on-investment targets. Most damningly, only about one in five AI decision-makers told Forrester that AI had improved their company’s operating profit over the past year. Money is flowing in at historic scale; profit is, for four out of five buyers, not flowing back. That is the demand-side mirror of Slok’s charts, and it sits alongside the build-out stories this site has covered, from Daytona’s agent cloud reaching a billion-dollar valuation to China’s plan to double its AI compute by 2027 — enormous energy on the selling side, hesitation on the buying side.
The other side: revolutions pay late
Slok’s critics have serious arguments, and the strongest one is history. General-purpose technologies have almost never paid off on the timetable of their promoters. Electrification took decades to show up in factory productivity, because the gains came not from swapping a steam engine for an electric motor but from redesigning the entire factory around electric power. Computers repeated the pattern: everywhere in the 1980s, the productivity surge arrived in the late 1990s. If AI follows the same arc, a flat 2026 tells us almost nothing about 2030, and Slok’s charts are a photograph of the installation phase, not a verdict on the technology.
There are two further counters. The first is that the Magnificent Seven’s gains may simply be the AI payoff, concentrated where the value currently pools: the firms selling picks and shovels in a gold rush genuinely are more productive, and their soaring revenue per worker is real economic value, not an accounting illusion. The second is measurement and price. Token prices have been falling steeply; a company can be using vastly more AI each quarter while its spending line — and the inference data Apollo cites — stays flat. Usage, revenue per worker and margins may all bend upward with a lag that quarterly data will only reveal in retrospect. Skeptics of Slok’s framing also note that splitting the index at exactly the seven biggest winners builds some of the conclusion into the test: any technology adopted unevenly will flatter its earliest, largest adopters.
Why this matters
This is not an academic quarrel about productivity statistics; it is an argument about the foundations of the stock market. A historically large share of the S&P 500’s value now rests on a chain of reasoning that runs: AI spending by the giants will continue, because AI is transforming their customers, whose rising profits will fund still more spending. Slok’s charts cut the middle link out of that chain. If the 493 are not seeing revenue or margin gains, then the hundreds of billions flowing into data centres are, for now, a transfer from the broad economy to seven companies — financed by investors who are assuming the transfer eventually reverses into broad-based profit. That assumption may prove right; the electrification precedent says late payoffs can be enormous. But an assumption carrying this much market capitalisation deserves to be measured rather than believed, and Slok has at least given the market a falsifiable test: watch the 493’s revenue per worker and margins, quarter by quarter. There is also a distributional edge: a technology whose gains accrue almost entirely to its sellers will reshape wage bargaining and white-collar hiring very differently from the broadly shared boom in the sales pitch.
What happens next
Slok has said, in his own framing, that “the AI story will be decided in the S&P 493, not the Mag 7” — and the calendar now supplies the evidence. Third-quarter earnings season is under way, and every results call from a bank, a retailer or a manufacturer is a data point in his experiment: do executives report AI-driven margin gains with numbers attached, or repeat the language of pilots and potential? Forrester’s 2027 projection will be tested as technology budgets are set this winter. If the 493’s lines stay flat through 2027 while spending keeps climbing, the burden of proof shifts decisively onto the bulls, and the repricing argument moves from a Daily Spark note to the centre of the market debate. If the lines bend — as they eventually did for electricity and for computing — Slok’s note will read the way the productivity paradox reads now: not wrong, exactly, but early. Either way, the question has been posed in the market’s own language, and it will not un-ask itself.