Yesterday, Torsten Slok, Apollo Global Management’s chief economist, published a blog post titled "No Signs of AI Payoff Outside the Tech Sector" befitting his role as a prominent "AI skeptic".
I asked ChatGPT (GPT-6 Pro) and Gemini (Argon High) to critique Torsten’s arguments. Both provided very good responses in my opinion; I would give Gemini an A and ChatGPT a B+. Below I excerpt the response from Gemini.
Gemini’s topline summary:
Torsten Slok's October 24, 2026 Daily Spark post, "No Signs of AI Payoff Outside the Tech Sector," is a classic example of the punchy, chart-driven macro commentary that has made Apollo Global Management’s chief economist one of Wall Street’s most widely read skeptics.
In just a few paragraphs and five charts, Slok argues that artificial intelligence has delivered "zero financial payoff" to corporate America outside of the Magnificent 7, that mainstream firms now view AI token spending as a margin-draining cost center, and that AI’s inability to exercise human judgment will structurally limit its bottom-line impact.
While Slok provides a much-needed empirical reality check against unbridled AI hype—particularly by exposing how heavily frontier AI demand relies on venture-capital recycling—his central conclusion conflates flat aggregate profit margins with zero technological utility and overlooks well-documented economic dynamics of how general-purpose technologies diffuse across the economy.
Gemini’s critique (Where Slok Overreaches: Methodological & Economic Blind Spots):
1. Confusing Profit Margins with Economic Payoff (Competitive Pass-Through). Slok assumes that if AI makes non-tech companies more efficient, their profit margins and inflation-adjusted revenue per worker must rise. Standard microeconomic theory says otherwise. In competitive industries (retail, commercial banking, airlines, logistics), when a cost-saving technology becomes widely available to all competitors, firms compete the cost savings away via lower prices, faster service, or higher quality to retain market share.
2. Ignoring the "Productivity J-Curve". Slok treats the current view of AI as a "cost center that weighs on margins" as proof of failure. However, economists Erik Brynjolfsson, Daniel Rock, and Chad Syverson have demonstrated that General Purpose Technologies (like electricity, computers, and cloud computing) follow a Productivity J-Curve. During the initial phase of adoption, firms must spend heavily on unmeasured intangible capital—cleaning proprietary data lakes, redesigning legacy workflows, retraining workers, and running parallel compliance tests…. Robert Solow’s famous 1987 quip that "you can see the computer age everywhere but in the productivity statistics"—made just a few years before the 1990s IT productivity boom began.
3. The Power-Law Fallacy in the Ramp Spend Data. In his fourth chart, Slok cites Ramp data showing that the top 10% of firms account for 99.5% of AI model-serving spend and more than 95% of OpenAI/Anthropic spend as evidence that AI is confined to a tiny sliver of the economy. Yet in that exact same chart, the top 10% of firms also account for 95.7% of all cloud computing spend. Nobody would argue that cloud computing has had "zero payoff" outside the tech sector.
4. Misinterpreting Model Commoditization. Slok notes that "mainstream corporate America does not need the latest model and is unwilling to pay for it," framing this as a sign of corporate retreat. In reality, this reflects healthy buyer rationality…. If a Fortune 500 insurance company switches from an expensive frontier reasoning model to a cheap, fine-tuned small model, its token spending drops while its operational utility stays high. That is a pricing-power problem for frontier AI labs and their VC backers—not evidence that the S&P 493 gets zero value from AI.
5. Aggregation Bias in the "S&P 493". The S&P 493 is a blunt instrument for isolating technological impact. It is heavily weighted toward physical, capital-intensive sectors—energy, utilities, real estate, materials, consumer staples, and heavy manufacturing—where macro variables like interest rates, commodity prices, healthcare medical-loss ratios, and tariff policies completely swamp the early margin impact of white-collar software tools.
It is interesting that Slok asserts confidently 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." I would ask Torsten whether in his judgment he would have benefited from having ChatGPT, Claude, and Gemini critique his blog post and provide additional context before he pressed publish.
