The meeting below never happened, and the people in it are illustrations; the decision it ends on is the one Torsten Slok spent 24 October documenting. A sales engineer for an AI platform has driven out to a regional food distributor to show its finance team what the newest model can do with their invoices and their call transcripts. They watch the demo through. They already run a cheaper, year-old model on a lower tier, and it clears the queue most days by lunch. When he names the price of the upgrade, the chief financial officer sets down her pen. The seat count stays where it is.

Slok is Apollo’s chief economist, and on that Saturday his Daily Spark note carried a flat verdict: “Despite massive investments and widespread hype around artificial intelligence, the technology has delivered zero financial payoff to corporate America outside of tech.” The sentence rests on charts, not on mood, and the charts divide the economy in two. On one side sit the Magnificent Seven, the handful of large technology firms building and selling the models. On the other is everyone else, the group Slok labels the S&P 493.

The charts that split the market in two

The first two charts track revenue per worker and profit margins. Both are rising for the Mag 7 and both are flat for the S&P 493, by Slok’s reading of Bloomberg and Macrobond data adjusted for inflation to the fourth quarter of 2019. The third chart carries most of the weight for the wider claim. It plots total factor productivity, the part of output growth left over once labour and capital are counted, and it shows no acceleration. In Slok’s words, “there are still no signs of AI boosting productivity growth.”

That productivity series is not Apollo’s own. It is the utilization-adjusted total factor productivity measure maintained at the Federal Reserve Bank of San Francisco, built on a method John Fernald set out in a 2014 working paper. The adjustment strips out the swings that come from firms running their machines and their people harder or softer through the cycle, so that what remains is closer to genuine technological change. For a general-purpose technology, that is the line to watch, and so far it has not bent.

From there Slok reads a behaviour. With nothing showing up 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.” The food distributor’s decision not to upgrade is that sentence in miniature. Mainstream firms, in his account, do not need the latest model and are unwilling to pay for it.

Half the economy has signed up and still cannot find the money

The puzzle sharpens against how widely the tools have spread. The Ramp AI Index, a monthly measure of business AI adoption kept by the economist Ara Kharazian at Ramp’s economics lab, put adoption at 55.9% of American companies in its September reading. More than half the firms in the country are paying for something. The payoff is missing anyway.

Where the money does concentrate, Slok finds a second worry. In a companion note two days earlier he cited Ramp data showing that 80% of the top 1% of AI spenders are venture-backed, against only 13% of the median spender. The heaviest buyers of compute are paying with investors’ capital rather than their own earnings. If the payoff is slow to arrive, he argued, venture returns come under pressure, and a pullback in funding would then feed back into weaker demand for compute, the same circularity observers have flagged in the deals between chipmakers, cloud providers and model labs, now visible one layer down in who settles the inference bill.

The same gap shows up in jobs and in the market

Slok’s note did not stand alone this month. Three other readings, taken with different instruments, pointed the same way.

The labour market is the first. Stanford’s Digital Economy Lab runs a set of public dashboards, launched in June and updated monthly, that track AI’s footprint on employment; in a 21 October write-up the university reported that employment for workers aged 22 to 25 in heavily exposed occupations such as software development and customer service has fallen, with a recent decline of 4.4%. The dashboards grew out of a paper by Erik Brynjolfsson, Bharat Chandar and Ruyu Chen that cast those young workers as canaries in the coal mine. The Federal Reserve heard the same thing from the other direction. Its October Beige Book recorded that “Five Districts reported that some firms had either reduced staffing levels (through attrition), delayed hiring, or anticipated less job creation after increasing productivity by investing in AI” — the first edition in which the survey’s standing sentence about AI described headcount coming down rather than tools going in.

Inside the firms that do see a payoff, the shift is sharper still. Anthropic’s own economists, in a 16 October report by Nathan Wilmers, Regina Catipon, Maxim Massenkoff and Peter McCrory, found that the proportion of junior technical hires at the company “fell even faster, from 27% to 9%,” and that by August its agents “worked 6 times as many hours as their human handlers.” That is what a payoff looks like where one exists: concentrated in a technology firm, and paid for partly in jobs that were never created.

The market supplied the fourth reading, and the least ambiguous. On 23 October Yahoo Finance’s Jared Blikre calculated that since the S&P 500’s record close on 6 October, “the companies building and equipping AI data centers are down a median 23%,” while “the Big Tech companies writing the checks are down just 5%.” The firms being paid fell four times as hard as the firms paying them, even as their order backlogs ran at record highs, a market betting that the spending slows before the returns show.

Why the flat line might hold

What Slok leaves open is whether the gap closes. Even as models grow cheaper and more alike, he asks whether mainstream firms can turn that cheap intelligence into higher revenue and fatter margins, and his own answer is cautious. “AI may be a great calculator,” he writes, “but in business, intelligence is mostly judgment.” Judgment depends on context and on who is asked. A model, he adds, “does not understand accountability, trust, culture, unwritten rules, ethical trade-offs or fairness,” the things that decide whether a customer stays and a deal closes.

That is an argument, not a measurement, and it belongs to him. The charts show a correlation holding flat; they do not prove AI can never move the S&P 493, only that it has not moved it yet. A single utilization-adjusted productivity series can miss gains that are real but slow to compound, and an adoption figure counts seats, not use.

Back at the food distributor, the finance team does the arithmetic it always does. The year-old model is good enough, the newer one costs more than the work it would save, and the quarter has to close either way. Somewhere inside the Mag 7, revenue per worker climbs another notch on Slok’s chart. The distance between those two facts is the whole of the story, and the next few quarters of the same chart will say whether it was a payoff arriving late or the shape of one that does not arrive at all.

References

  • Blikre, J. (2026). The AI sell-off is crushing stocks of companies leading the data center build-out. Yahoo Finance. finance.yahoo.com
  • Brynjolfsson, E., Chandar, B., & Chen, R. (2026). Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence. Stanford Digital Economy Lab. digitaleconomy.stanford.edu
  • Federal Reserve System (2026). Beige Book, October 2026: National Summary. federalreserve.gov
  • Fernald, J. (2014). A Quarterly, Utilization-Adjusted Series on Total Factor Productivity. Federal Reserve Bank of San Francisco. frbsf.org
  • Hadhazy, A. (2026). Tracking AI’s Real-Time Footprint on the Labor Market. Stanford HAI. hai.stanford.edu
  • Kharazian, A. (2026). Ramp AI Index. Ramp Economics Lab. ramp.com
  • Slok, T. (2026). Circularity in AI Shows Up in Token Spending Too. Apollo, The Daily Spark. apollo.com
  • Slok, T. (2026). No Signs of AI Payoff Outside the Tech Sector. Apollo, The Daily Spark. apollo.com
  • Wilmers, N., Catipon, R., Massenkoff, M., & McCrory, P. (2026). How employment is changing at Anthropic. Anthropic. anthropic.com