Apollo Global Management's chief economist published a short note on 24 October that reads like a cold shower for the enterprise AI trade. Torsten Slok looked at what he calls the S&P 493 — the S&P 500 minus Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla — and found that revenue per worker is rising for the seven giants and flat for everyone else. Profit margins tell the same story: up for the Magnificent Seven, flat for the rest. And on total factor productivity, using the San Francisco Fed's utilisation-adjusted series, there is still no sign that AI is lifting the economy's underlying efficiency.
The note, titled "No Signs of AI Payoff Outside the Tech Sector", also draws on Ramp's AI spending data for a second observation: AI spending is extremely concentrated in a small number of firms, and token spending displays the same circularity that critics have flagged in chip deals, with the largest buyers often selling to each other.
Slok's conclusion is behavioural as much as statistical. With no impact on the bottom line, he writes, "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." Venture-backed companies keep buying the newest model. Ordinary companies, in his account, do not need it and are unwilling to pay for it.
He is equally sceptical that cheap intelligence converts into profit on its own. "AI may be a great calculator, but in business, intelligence is mostly judgment," he writes, and judgment "depends on context". He then lists what he believes models do not grasp: accountability, trust, culture, unwritten rules, ethical trade-offs and fairness — "all of which shape reputational risk and ultimately drive revenue growth and profit margins."
The framing is the sharpest part of the argument. "The AI story will be decided in the S&P 493, not the Mag 7," Slok writes, "and so far, there is no evidence that revenue growth or margins are improving outside the tech sector." That inverts the usual reading of the boom: the seven companies can report spectacular results indefinitely and still leave the productivity case unproven, because their customers are the ones who are supposed to absorb the tools and convert them into output.
What the note does not do is settle the question. It is a research commentary, not a peer-reviewed study, and margin data is a lagging indicator — general-purpose technologies typically take years to show up in aggregate statistics, which is precisely why the productivity debate has run for three years without resolution. Slok's own charts are indexed series whose exact construction sits behind Apollo's data licences, and a seven-company comparison is a blunt instrument for a technology that may diffuse unevenly by sector and firm size.
The counter-evidence is not trivial either. AWS chief executive Andy Jassy said this week that 2026 cloud capacity is sold out, that 2027 is largely sold out and that "a good chunk" of 2028 is already committed, with Amazon's 2026 capital spending near $220 billion. Somebody is buying. The open question is whether the buyers are a narrow ring of labs and their venture-funded customers — which is Slok's point — or the beginning of broad adoption that simply has not reached the income statement yet.
Markets get a partial answer this week, with Microsoft reporting on Wednesday and the rest of big tech following. Watch two numbers in particular: whether non-tech companies in the S&P 493 begin disclosing AI-related cost lines rather than savings, and whether the margin gap between the Mag 7 and the rest widens further. Slok's thesis is falsifiable, which is more than can be said for most AI forecasts. If margins outside tech start climbing in the next two quarters, the note ages badly. If they stay flat while capex keeps rising, the burden of proof shifts decisively onto the companies selling the boom.
