Apollo says AI has produced zero financial payoff outside tech sector

Apollo Global Management chief economist Torsten Slok published a note on Oct. 24, 2026, finding that artificial intelligence has delivered zero financial payoff to corporate America outside the technology sector.

The headline finding is contested. Remio cites survey evidence of modest productivity benefits, complicating the claim that AI has produced no payoff outside tech. Remio characterizes Apollo’s note as interpretive rather than results from a controlled study of enterprise rollouts.

Revenue per worker climbs for the Magnificent Seven but stays flat for the S&P 493, according to Slok’s note. This is the core metric behind Apollo’s zero-payoff finding. Profit margins climb for the Magnificent Seven but remain flat for the S&P 493, according to the same note. This is the second core metric in Apollo’s comparison.

The Magnificent Seven companies invest heavily in AI, and Microsoft, Meta, Alphabet, Amazon and Nvidia sell significant volumes of AI processors, cloud resources and related infrastructure. The group of 493 non-tech S&P 500 companies is not a pure sample of firms outside the technology sector, and the technology giants are not a clean experiment in AI adoption.

Slok’s comparison fails to isolate AI’s effect at individual companies and does not prove AI caused the gap between the Magnificent Seven and the other 493 companies. Both S2 and Remio note the limits of Apollo’s causal inference.

Mainstream corporate America is retreating and viewing AI token spending as a cost center that weighs on margins rather than a source of growth, according to Slok, describing the shift in enterprise attitude toward AI spending. Venture capital-backed companies are spending more on tokens while mainstream corporate America does not require the latest model and will not pay for it, a distinction between VC-driven AI spending and mainstream enterprise reluctance.

Apollo argues that macroeconomic effects of AI will remain very limited if agentic AI and token spending stay confined to a narrow slice of the economy, Apollo’s boundary condition for broader AI economic impact.

Apollo’s note argues that AI does not understand accountability, trust, company culture, informal rules, ethical trade-offs or fairness. Apollo’s argument holds that these gaps feed reputational risk and affect revenue growth and profit margins, which is why the note expects limited bottom-line impact outside technology. Apollo says the bottom line will be decided in the S&P 493 rather than the Magnificent Seven, and no evidence yet shows revenue growth or margins improving outside the technology sector, Apollo’s concluding framing of the AI payoff debate.

The chief economist said, “Revenue per worker is rising for the Mag 7 but is flat for the S&P 493.” The finding has also been summarized as “zero payoff” by Torsten Slok, Apollo Global Management chief economist, as reported by Bloomberg.

A firm AI adoption study revised in October 2026 surveyed almost 9,000 senior executives in seven countries and found that 78% of surveyed firms use AI, directly contesting Apollo’s literal zero-payoff claim by showing broad adoption and reported benefits. Executives in that survey said AI adoption added a 0.2% annual increase to labor productivity growth over the previous three years and projected a bigger 0.6% yearly contribution over the next three years, quantifying modest reported productivity benefits from AI adoption. According to the survey authors Ivan Yotzov, Jose Maria Barrero, Nicholas Bloom and collaborators, concentrated spending means just 2% of firms make up over half of reported AI expenditure, showing AI spending is concentrated in a small share of firms. Fresh survey evidence undermines a literal claim of zero payoff while leaving widespread financial transformation unproven, Remio’s synthesis of why Apollo’s headline is too strong.

Erik Brynjolfsson, Danielle Li and Lindsey Raymond studied a conversational assistant deployed among 5,172 agents and found that issues resolved per hour rose by an average of 15%, a concrete example of task-level AI productivity gain in customer support. Gains in that support study varied, with less experienced and lower-skilled agents improving speed and quality while top performers received smaller speed gains and small quality drops, showing the uneven distribution of AI benefits across worker skill levels.

A workplace AI experiment by Eleanor Wiske Dillon, Sonia Jaffe, Nicole Immorlica and Christopher Stanton studied 7,137 knowledge workers across 66 firms and found that active users cut weekly email time by two hours in the later period, an example of worker-level time savings without broader output changes. The researchers found no broader changes in what tasks workers did or how much work they completed from individual access to the AI assistant, the limits of individual AI access in the workplace experiment.

Danish researchers Anders Humlum and Emilie Vestergaard linked adoption surveys with administrative labor-market records and found widespread changes in work without detectable changes in pay or recorded hours, international evidence that AI adoption has not yet shown up in earnings or hours data.

In the four quarters through the second quarter of 2026, unadjusted total factor productivity, a measure of economic output relative to combined inputs, increased by 1.42%, while its utilization-adjusted counterpart declined by 0.07%, illustrating how productivity measurement choices affect the AI debate. Saved time turns into a financial return only if an organization can use that time or capture another measurable benefit, explaining the conversion problem between AI adoption and financial ROI.

Accelerated output at an unconstrained stage may leave overall delivery unchanged when downstream steps such as security review, integration testing and release approval remain bottlenecks, relevant to CISO scrutiny of AI-driven security tooling and workflow integration. Finance teams must separate a promising demo from an operating result, since the latter shows whether applying a tool changes the performance of a real workflow, guidance for enterprise buyers evaluating AI-security products.

Microsoft’s latest quarterly capital spending totaled $41 billion, an increase of 70% from the previous year, part of hyperscaler AI infrastructure spending context. Microsoft’s calendar-year 2026 spending guidance would reach roughly $175 billion due to an accounting change, though the underlying investment plan stays unchanged, the company said, showing the scale of planned AI-related capital investment by a major tech supplier.

Meta raised its 2026 capital expenditure guidance to a range of $135 billion to $145 billion, illustrating aggressive AI spending by a non-supplier tech giant. U.S. hyperscalers are expected to spend about $800 billion in total capital during 2026 according to consensus estimates cited by Goldman Sachs, though not all of that outlay is exclusively for AI, an aggregate infrastructure spending backdrop for the AI ROI debate.

Microsoft, Meta, Alphabet, Amazon and Nvidia sell significant volumes of AI processors, cloud resources and related infrastructure, and all seven Magnificent Seven companies invest heavily in AI, identifying the supply-and-demand concentration in AI infrastructure.

CoreWeave‘s five-year credit default swap spread, a market gauge of default risk, touched a record high of 1,194.5 basis points, or hundredths of a percentage point, on Oct. 15, implying roughly a 63% chance of default, illustrating debt-market stress in AI infrastructure financing. Oracle’s credit default swap spread climbed near 300 basis points during a $466 billion AI borrowing surge, further evidence of leverage concerns around AI infrastructure buildout. At the end of June, CoreWeave, which is backed by Nvidia, carried roughly $35 billion in debt and about $16 billion in debt-like lease obligations, detailing the company’s leverage position.

Shares of Nvidia-backed CoreWeave dropped more than 15% on Oct. 14 after venture capitalist Chamath Palihapitiya highlighted that yields on $1.25 billion of bonds due in 2032 briefly reached 15%, describing a mid-October market scare over AI financing. Yields on CoreWeave’s unsecured bonds subsequently remained near 15%, and Oracle plus several other AI infrastructure equities also experienced a sell-off, showing the persistence of financing stress after the initial shock.

Market research firm Gartner estimated that AI infrastructure spending worth more than $3.5 trillion could be impaired by 2030 should AI revenue fail to justify the outlay, quantifying potential impairment risk if AI revenue does not materialize. Meta said it plans to increase AI wagering and capital expenditure, and the reaction pushed its shares down while its credit risk gauge reached a record high, illustrating investor jitters over AI capital spending.

Australian data center company Firmus Grid Ltd. saw its initial public offering plans collapse after U.S. fund managers objected to the $30 billion valuation, an example of valuation pushback in AI infrastructure IPO market. Nscale Ltd., another Nvidia-backed data center developer, faced pushback over its own IPO valuation, another sign of skepticism toward AI infrastructure valuations.