Apollo's Slok: AI Has Paid Off Only for the Mag 7, Not the S&P 493

Apollo Global Management's chief economist, Torsten Slok, argued in a note published on Saturday, 24 October 2026, that AI has produced no measurable financial payoff for large US companies outside technology. TechCrunch reported that he looked at the S&P 500 minus Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla, a group he calls the "S&P 493". He found profit margins and revenue per worker flat and no evidence of a productivity gain. For executives approving 2027 AI budgets, it is a ready-made question from the board: where is the return?

What exactly did Apollo's Torsten Slok say?

The claim is narrow and comes from a short chart-based note. Bloomberg quoted him as writing that revenue per worker is rising for the Mag 7 but is flat for the S&P 493. He said the same is true of profit margins, and that there are no signs yet of AI boosting productivity growth.

His conclusion, as relayed by AI Weekly, is that the AI story "will be decided in the S&P 493, not the Mag 7." He is sceptical that non-tech firms can turn cheaper intelligence into higher revenue growth and margins. Apollo's Daily Spark page lists the note's three charts: revenue per worker, profit margins and productivity growth.

TechCrunch notes that Apollo is the private equity firm that used to own TechCrunch. That is a disclosure, not a reason to doubt the report. The Bloomberg text we could access confirms the quotes independently.

Do the numbers hold up?

Partly, and the coverage does not agree on the details. Implicator describes the first chart as showing Mag 7 revenue per employee, adjusted for inflation, rising from about $550,000 in 2022 to almost $800,000 in 2026. It puts the S&P 493 at around $550,000 since 2019. Cryptopolitan gives about $780,000 for the Mag 7 and near $525,000 for the rest.

We could not view the charts, and Apollo's page carries only chart titles. The gap is probably a matter of reading values off a graph. The direction is the same in every account we read: one group is pulling away and the other is not.

Cryptopolitan also reports margins of close to 30% for the seven giants and about 10% for the rest. We found that in no other source, so treat it as unconfirmed. The same goes for the claim, repeated by AI Weekly and Cryptopolitan, that the top 10% of firms account for 99.5% of model-serving and inference spending. Neither outlet links it to a document we could check.

There is a larger caveat. These are aggregate company numbers, and they do not isolate AI's effect. Flat averages fit with AI doing nothing. They also fit with gains in some firms or functions that are offset by costs, price cuts or weaker demand elsewhere. A note of five charts cannot separate those explanations.

Is AI adoption outside tech actually stalling?

Not by the usual adoption counts, but spending growth has slowed. Ramp's AI Index, based on transactions from more than 85,000 US businesses, shows 55.9% paying for AI in September. That is up 0.6 points from August and 12.1 points from a year earlier. Retail rose 1.6 points to 49.9% and construction rose 1.9 points to 43.2%.

Ramp's lead economist, Ara Kharazian, wrote in an October 14 update that growth in new AI customers has slowed a lot and the market looks to be at the top of the S-curve. He added that further growth will have to come from more intense use rather than more customers. The same piece records two straight months of virtually no growth in token spend, which Apollo's note draws on (a secondary summary says it cites Ramp for two of its five charts).

Ramp's sample is its own customers, so it skews toward companies that use a spend-management platform. The Census Bureau's survey gives a more conservative view. A tracker built on the Bureau's biweekly data reports 23.8% of US businesses used AI in the two weeks ending 4 October 2026. A Census working paper found AI use concentrated in large firms and knowledge-intensive sectors, with labour declines rare.

Why cheaper tokens complicate the story

Falling prices can make spending data look weak even when use is rising. A Substack analysis of the Ramp data says Ramp's index of the effective price per million tokens fell from $1.15 in March to $0.68 in September. It says the same dollar bought roughly 70% more tokens. It also reports that frontier models' share of token volume on Ramp fell from a 53% peak in August to 45%, as companies trade down to cheaper models.

TokenPost reports that Ramp's latest weekly data showed token spending up 23.0% from a month earlier, ending the two-month stall. It also reports that the top 1% of businesses in Ramp's sample account for 80% of combined OpenAI and Anthropic spending. Taken together, heavy use is concentrated in a small group of firms, which fits Slok's sceptical reading of the broad economy. It is also consistent with a slow, uneven rollout.

For vendors, this matters more than the headline. Prices per token are falling, so revenue growth has to come from volume or from higher-value workflows. A buyer that sees no margin change will push back on both.

What should operators do with this?

Do not treat the note as a verdict. Treat it as the argument your finance team will bring to the next budget meeting. Three practical steps follow from the evidence above.

• Measure outcomes at the process level. Company-wide revenue per worker moves too slowly to show a pilot's effect. Track cycle time, cost per ticket, conversion or error rates in the specific workflow that changed.
• Separate adoption, depth and results. Ramp's own economist says adoption is a weakening headline metric once more than half of businesses pay for AI. Report seats, usage intensity and business outcomes as three different numbers.
• Budget on unit economics. With token prices falling, flat spend can mean more usage. Compare cost per completed task, not total invoice.

What to watch next

The next test is earnings. An AI Weekly commentary argues that if S&P 493 margins stay flat through fourth-quarter earnings season, pressure on hyperscaler guidance and private AI valuations would build. That is the commentator's inference, not Apollo's forecast. Still, it names the right indicators: margin commentary from non-tech companies, and whether Ramp's token spend keeps rising after the late-October rebound.

Also watch the Census series, which updates every two weeks. A move from the mid-20s toward 30% would show breadth is arriving. If it stalls while Ramp's heavy users keep spending, the split Slok describes will look structural rather than temporary.