The AI payoff outside tech is still missing, Apollo Global Management chief economist Torsten Slok argued in a Saturday note: revenue per worker and profit margins are rising for the Magnificent Seven and flat for the other 493 companies in the S&P 500. He sees no clear productivity lift from artificial intelligence, and says mainstream firms are starting to treat token spending as a margin drag rather than a growth bet. For enterprise leaders outside the largest technology platforms, the claim reframes AI spend as a return-on-investment problem, not an adoption race.
Key Takeaways
- Revenue per worker is rising for the Mag 7 but flat for the S&P 493.
- Mag 7 profit margins are near 30%; S&P 493 margins are around 10%.
- The top 10% of firms account for 99.5% of inference spend.
- More than $3.5 trillion of AI infrastructure spending could be impaired by 2030 if AI revenue fails.
The AI Payoff Outside Tech Still Looks Flat
Slok’s Saturday note, also posted on Apollo’s Daily Spark, draws a hard line between the platform companies and everyone else. Both revenue per worker and profit margins are rising for the Mag 7 and flat for the S&P 493. He finds no signs yet that AI is lifting productivity growth. The levels behind the charts, as reported by Stocktwits, are in the table below.
Apollo’s note on the revenue-per-worker chart says that series is inflation-adjusted to the fourth quarter of 2019, using Bloomberg and Macrobond. The margin chart lists Bloomberg and Macrobond but does not say the series is inflation-adjusted. Stocktwits separately describes the revenue-per-employee figures as inflation-adjusted.
| Measure | Magnificent 7 | S&P 493 |
|---|---|---|
| Revenue per employee (inflation-adjusted) | Roughly $780,000 | Around $525,000 |
| Profit margins | Nearly 30% | Around 10% |
| Recent direction, per Slok | Rising | Flat |
“Revenue per worker is rising for the Mag 7 but is flat for the S&P 493,” he wrote, in the account carried by Yahoo Finance, citing Bloomberg. Yahoo also says the same pattern holds for profit margins. On Apollo’s site he went further, writing that the technology “has delivered zero financial payoff to corporate America outside of tech.”
The productivity chart uses utilization-adjusted total factor productivity from the San Francisco Fed — the Fernald series, adjusted for cyclical variation in factor use — with Bureau of Labor Statistics data. Slok reports no clear acceleration, even after the surge in AI investment.
Token Spend Is Turning Into a Cost Center
“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,” Slok said. Venture-backed firms are spending more on tokens. Mainstream companies, he argues, do not need the latest model and are unwilling to pay for it.
Spending is tightly concentrated, according to figures Stocktwits drew from Slok’s note and from Ramp chief economist Ara Kharazian and the Ramp AI Index:
- The top 10% of firms account for 99.5% of model serving and inference spend.
- That group accounts for 99% of neocloud spend and more than 95% of spending tied to OpenAI and Anthropic.
- In cloud computing, the top 10% account for 95.7%.
- Among the top 1% of AI spenders, 80.4% are venture-backed, versus 13.1% among median AI spenders.
“As long as agentic AI and token spending are confined to a small part of the economy, the macroeconomic effects of AI will be very limited,” he wrote.
Infrastructure Bets Meet a Sharper ROI Test
Meta kicked off tech results last week by saying it intends to bet even more aggressively on AI and boost capital-spending plans. That sent its shares lower and its credit-risk gauge to a record high, Bloomberg reported via Yahoo Finance. Comments from CoreWeave Inc. accentuated worries over the extent to which the AI infrastructure buildout is being financed by debt.
Capital markets are already pushing back on data-center valuations. IPO plans for Australian data center company Firmus Grid Ltd. collapsed after US fund managers balked at the $30 billion valuation. Nvidia-backed developer Nscale Ltd. has faced pushback over its own IPO valuation.
Gartner estimated that more than $3.5 trillion of AI infrastructure spending could be at risk of impairment by 2030 should AI revenue fail to justify the spending.
Why the S&P 493 Decides the Enterprise Case
“The AI story will be decided in the S&P 493, not the Mag 7,” Slok said, Stocktwits reported. The question he poses is whether mainstream corporate America can turn increasingly cheap intelligence into higher revenue growth and higher profit margins even as models become commoditized. He writes that may be harder than it sounds.
On Apollo’s site he writes that AI “may be a great calculator, but in business, intelligence is mostly judgment.” Judgment “depends on context, and the answer varies with which human or which model you ask.” He adds that AI “does not understand accountability, trust, culture, unwritten rules, ethical trade-offs or fairness,” factors that shape reputational risk and ultimately drive revenue growth and profit margins.
For large organizations, the practical reading is analysis, not a new data point: AI programs that cannot be tied to revenue per worker or margin will look like overhead once finance reviews the token bill. The same cost logic sits behind why sold-out cloud capacity keeps inference expensive.
