News · 2026-10-25

Apollo sees no broad AI payoff outside tech, but its charts cannot isolate AI’s effect

Apollo chief economist Torsten Slok says AI’s financial payoff is not showing up broadly outside major technology companies. His October 24 analysis compares the Magnificent Seven with the other 493 S&P 500 companies, finding stronger revenue per worker and margins among the giants. The charts make a useful challenge to the investment story, but do not establish that AI caused either group’s performance.

Key facts

• Apollo compares seven large technology companies with the remaining 493 S&P 500 companies.
• Slok published the analysis on October 24, 2026.
• The indicators include revenue per worker, profit margins, and utilization-adjusted total factor productivity.
• The primary source is Apollo’s Daily Spark analysis.

The striking part of this argument is where Slok looks for the return. A supplier can sell more AI services without its customers becoming more profitable. If software companies and chipmakers prosper while ordinary businesses simply add another subscription bill, investors have evidence of spending, rather than evidence of a widely shared productivity transformation.

Slok writes that AI has “delivered zero financial payoff” outside tech. That is the economist’s interpretation, not a measured treatment effect. His comparison is a dashboard: it tells readers which broad indicators appear to be moving together. It does not construct the counterfactual world in which the same businesses made the same investments but had no AI tools.

That distinction matters because revenue per employee can change for several reasons. A company may raise prices, change its product mix, reduce headcount, or grow faster than it hires. A rising margin can reflect cost control or market power. Conversely, an AI tool can improve one task while its savings disappear into implementation costs, customer discounts, or other parts of the organization. Those are possible explanations, not findings established by Apollo’s charts.

The inflation-adjusted revenue-per-worker chart uses a 2019 fourth-quarter price reference. Apollo credits Bloomberg and Macrobond for company data, and its productivity chart draws on a San Francisco Federal Reserve series and the Bureau of Labor Statistics. The visible text does not provide the chart window, a precise margin definition, or a method that assigns changes to AI. The S&P 493 also does not represent every nontechnology business, particularly small firms and private companies.

An important counterweight comes from the Bureau of Labor Statistics’ revised second-quarter release. It reports nonfarm-business labor productivity rising 1.4% at an annualized rate from the first quarter and 2.2% over a year. That is evidence of productivity growth, but the bureau does not attribute it to AI. Labor productivity and utilization-adjusted total factor productivity are different measures, so treating one as a direct rebuttal of the other would compound the confusion.

The strongest argument against drawing a pessimistic conclusion is that adoption may require complementary work. A faster drafting tool does little for revenue when approvals, data access, or delivery remain slow. Participants in the Hacker News discussion point to workflow redesign and the unusually testable nature of coding as reasons software businesses may see gains earlier. Those are plausible mechanisms and reader interpretations, not independent measurements of the payoff.

Labor-market evidence needs the same discipline. In a separate, September 27 Apricitas Economics essay, Joseph Politano estimates more than 200,000 net job losses across selected creative industries over four years. He does not supply a number attributable to AI. Film-industry strikes, consolidation, offshoring, and shifts toward social platforms are among the other pressures he discusses. The bureau’s motion-picture employment account independently identifies strike-era disruption.

For a business reader, the useful response is to ask for outcomes at the right level. A demonstration can show that an employee completes a task faster. A company case study can show whether that saving survives integration and review. A broad productivity series asks whether such gains become visible across the economy. Moving between these levels requires evidence; a persuasive anecdote cannot do the work of all three.

Ground Truth’s lessons on randomized trials and causal graphs explain what stronger attribution would require. A stronger company-level case would track completed work, error rates, review time and implementation costs together, rather than measuring only how quickly a first draft appears. Apollo supplies a timely demand for that evidence. Its honest limit is that a flat aggregate comparison cannot prove that every AI investment failed, just as rising national productivity cannot prove that AI succeeded.

Primary source, verified: read the paper →

Ground Truth. (2026, October 25). Apollo sees no broad AI payoff outside tech, but its charts cannot isolate AI’s effect. Ground Truth. https://groundtruth.day/news/apollo-sees-no-broad-ai-payoff-outside-tech.html

@misc{groundtruth:apollo-sees-no-broad-ai-payoff-outside-tech,
  title  = {Apollo sees no broad AI payoff outside tech, but its charts cannot isolate AI’s effect},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {oct},
  url    = {https://groundtruth.day/news/apollo-sees-no-broad-ai-payoff-outside-tech.html}
}

Topics: economics · productivity · industry · ai-adoption