NextFin News - Artificial intelligence has delivered "zero financial payoff" to corporate America outside of tech, despite the massive investments and widespread hype surrounding the build-out, according to Torsten Slok, chief economist at Apollo Global Management. Revenue per worker and profit margins are rising for the Magnificent Seven but flat for the other 493 companies in the S&P 500, and there are still no signs that AI is lifting productivity growth, Slok wrote in a note published Saturday. The verdict reframes the AI boom as a narrow concentration trade rather than a broad-based productivity shock — and it sets up a direct confrontation between the earnings the market has priced and the earnings the broader index is actually producing.
The stakes are unusually high because the two claims cannot both be comfortably true for long. Either the S&P 493 begins converting cheap intelligence into revenue growth and fatter margins, validating a stock market priced for an AI-driven earnings expansion, or the spending cycle that is currently supporting a large share of index profits slows under its own weight. Slok's answer to which path is more likely is unambiguous: "The AI story will be decided in the S&P 493, not the Mag 7, and so far, there is no evidence that revenue growth or margins are improving outside the tech sector."
The Evidence: A Two-Speed Economy
Slok's case rests on three charts drawn from market and macroeconomic data compiled in the note. Inflation-adjusted revenue per worker — a clean proxy for whether technology is making the average business more productive — is climbing for the Mag 7 while it is flat for the S&P 493. Profit margins tell the same two-speed story. The seven mega-cap technology leaders have pushed margins from roughly 15% in the first quarter of 2023 to about 25% in early 2026, a 10-percentage-point expansion that captures the pricing power and scale economics of selling AI. Margins for the rest of the index have hovered near 10% over the same stretch, and the broader 500-company gauge has held around 12%.
The third chart is the one that should trouble investors underwriting a productivity-led earnings expansion. Using the San Francisco Federal Reserve's utilization-adjusted total factor productivity series — the standard measure of how efficiently labor and capital are combined after stripping out the business cycle — there is no detectable pickup attributable to AI. The series, adjusted for cyclical variation in factor utilization, has not accelerated. If AI were diffusing through the corporate base as a general-purpose technology, the first place it would show up is exactly where it has not: output per worker and margins for the 493 companies that make up roughly two-thirds of the index by count.
"With no impact on the bottom line, 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," Slok wrote.
The concentration is not limited to earnings; it is baked into the index itself. The Mag 7 accounted for 34.6% of S&P 500 market capitalization as of October 23, 2026, up from 19.5% at the end of 2022 and 13% at the end of 2018 — a level of single-theme concentration not seen since the Nifty Fifty era of the early 1970s. Yet in 2026 the performance gap has flipped. By mid-2026, the S&P 493 had contributed roughly 96% of the index's price-only return for the year, with six of the seven mega-cap leaders in double-digit drawdowns from their peaks. The market is simultaneously assigning a third of its value to seven companies while drawing almost all of its 2026 gains from the other 493.
Why the Payoff Is Confined: Intelligence Is Not Judgment
The first-order explanation for the divergence is simple accounting: the Mag 7 are the sellers of AI, and the S&P 493 are the buyers. Hyperscaler capital expenditure is on track to reach roughly $800 billion in 2026, and AI-infrastructure beneficiaries are expected to account for about half of S&P 500 earnings growth this year. That spending flows through the earnings of chipmakers, hardware suppliers, utilities, and the hyperscalers themselves. It does not, by construction, show up as margin expansion at a grocer, an insurer, or a manufacturer — the buyers record it as an expense line while the sellers record it as revenue.
But the deeper question is why adoption has not converted cheap intelligence into profit even where the tools are available. Slok's answer is that AI's comparative advantage is computation, while business value is mostly judgment. "AI may be a great calculator, but in business, intelligence is mostly judgment," he wrote. "Judgment depends on context, and the answer varies with which human or which model you ask." Most importantly, he argues, "AI does not understand accountability, trust, culture, unwritten rules, ethical trade-offs or fairness, all of which shape reputational risk and ultimately drive revenue growth and profit margins."
That distinction maps onto the deployment friction that has shown up in company after company. A study published last year found that only 5% of companies saw a meaningful return from generative AI pilot projects, meaning 19 out of 20 pilots failed to clear the hurdle that would make them more than experiments. Ford Motor hired 350 veteran engineers to retrain junior staff and reprogram AI tools that were not effective without human oversight, even as it deployed AI vision systems across 33 global plants. Ricoh outsourced insurance-claims administration to AI and ended up paying roughly three times the manual cost — about $500,000 in consulting fees plus $200,000 a month in AI fees — while cutting headcount only from 44 to 39. The productivity gain was real, but it arrived slowly and expensively.
This is the mechanism behind the flat revenue-per-worker line. The payoff requires business-process redesign, data governance, regulatory clearance, and human-in-the-loop verification. Those are one-time implementation costs that hit margins before the benefit arrives, and in many workflows the benefit is smaller than the pilot suggested because the tasks that are easiest to automate are also the ones that contribute least to top-line revenue. The result is a J-curve that many companies are still descending: costs rise immediately, benefits accrue only after the organization has been rewired, and for a large share of firms the rewiring never happens because the processes that would benefit most are also the ones entangled with judgment.
Slok's phrasing captures the asymmetry: "VC-backed companies are spending more and more on tokens, but mainstream corporate America does not need the latest model and is unwilling to pay for it." The companies racing to consume the most intelligence are the ones whose business model is to resell it; the companies that would need to absorb it into existing operations face a much harder integration problem with a less certain payoff.
The Circularity: Earnings Growth That Depends on Spending Growth
The second-order problem is that the S&P 500's current earnings strength is partly circular. Research from Goldman Sachs estimates that almost half of S&P 500 earnings-per-share growth in 2026 comes from AI investment itself. The spending is the earnings — for the chipmakers building the accelerators, the utilities powering the data centers, the equipment suppliers, and the hyperscalers booking the revenue. That creates a chain in which today's reported profit growth is funded by today's capital outlays, and those outlays are justified by the expectation of future profit growth that has not yet appeared anywhere outside the technology sector.
The same loop shows up on the demand side, and it is even more concentrated. Among firms paying for AI, the top 1% of spenders account for the overwhelming majority of token consumption. In the week ending October 11, 2026, Anthropic captured 60.1% of business token spending and OpenAI 36.4%, leaving 3.5% for every other provider combined. The top 1% of AI spenders were paying a median of $9,567.53 per employee per month, compared with $15.70 for the median firm — a 600-to-1 spread that shows adoption is broad by headcount but narrow by dollars. As long as agentic AI and token spending are confined to a small part of the economy, Slok writes, "the macroeconomic effects of AI will be very limited."
The circularity creates a fragility that becomes visible under a mean-reversion lens. If the S&P 493 does not start converting intelligence into margin, token-spending growth slows. If token-spending growth slows, hyperscaler revenue growth slows. If hyperscaler revenue growth slows, the capital-expenditure program that is currently supporting roughly half of index earnings growth comes under review. The chain does not require a collapse in AI's usefulness — only a deceleration in the rate at which spending accelerates. That is a lower bar than most AI-bear arguments require, and it is why the payoff question is a market question, not just a technology question.
The balance-sheet footprint of the build-out underscores the risk. Consensus estimates imply the seven largest technology companies will spend about $770 billion on capital expenditure in 2026, equivalent to 100% of their cash flow from operations. To keep funding the program, they have increasingly turned to debt and equity issuance and pulled back on buybacks; net debt for the hyperscalers has risen by $170 billion since the start of 2025, and share counts have begun to increase. At the same time, rising depreciation from the build-out is expected to reverse some of the return-on-equity benefit that AI capex generated through elevated semiconductor margins. Consensus implies that ROE for the seven largest technology stocks will decline by an average of 700 basis points next year, even as the aggregate S&P 500 sits at a record 22% trailing ROE.
The Counter-Thesis: Surplus, Not Profits, Is the Point
The strongest challenge to Slok's framing is that it measures the wrong thing. Economist J. Bradford DeLong, responding to the note, accepts the narrow comparison but argues that the gains from a general-purpose technology show up first as consumer and business surplus — cheaper services, faster problem-solving, new firms — rather than as measured GDP or incumbent profits. On this view, the absence of a profit explosion in the S&P 493 does not falsify the AI thesis; it is the expected transitional phase before the new businesses that capture the surplus show up in the index.
There is also a timing argument, and it is backed by numbers. Goldman Sachs embeds only a 0.4 percentage point boost to S&P 500 EPS growth from AI-driven productivity in 2026, rising to 1.5 percentage points in 2027 — a deliberately gradual ramp that assumes the payoff arrives later, not never. Adoption is also broadening: 55.9% of U.S. businesses on Ramp's platform paid for AI in September 2026, up 0.6 percentage points from August and 12.1 percentage points from a year earlier. The bull case is that diffusion is a lagging indicator of profit impact, and that the margin expansion Slok is waiting for is already in the pipeline.
The counter-thesis is coherent, but it carries its own burden. Surplus that does not accrue to corporate profits cannot support equity valuations that are priced on profit growth. The S&P 500 trades at about 21 times forward earnings, in the 87th percentile since 1980, and index return on equity sits at a record 22%. Those multiples are being underwritten by an earnings trajectory that requires the S&P 493 to eventually do what it has not done for three and a half years: turn AI into margin. If the surplus argument is right but the profits never follow, the index multiple has priced a payoff that accrues to users rather than shareholders — a mismatch that resolves through price, not through earnings.
Scenarios and What to Watch
The debate is resolvable, and the falsifying signal is concrete. Slok's "zero payoff" thesis would be wrong if two things happen together over the next four quarters: utilization-adjusted total factor productivity accelerates above its 2010–2019 trend, and S&P 493 profit margins expand by at least 100 basis points. Either one alone could be noise; both together would mean diffusion has begun, and the market's current pricing would look prescient rather than stretched.
Three scenarios frame the path from here:
- Base case — a long ROI runway with a mid-cycle air pocket. The S&P 493 eventually converts AI into modest margin gains, but not fast enough to justify the capex trajectory embedded in 2026 earnings. Hyperscaler spending growth decelerates from the rapid pace of 2025 toward the roughly 20% consensus expects for 2026, depreciation weighs on mega-cap ROE, and the index works off valuation through a period of flat returns rather than a sharp collapse. In this world the payoff is real but slow, and the market pays for the optimism with time.
- Downside case — the circularity breaks. Token-spending growth stalls before the S&P 493 delivers, the capex program is cut back, and the earnings contribution from AI infrastructure reverses faster than expected. With the seven largest technology stocks carrying a collective 44% ROE and a 34.6% index weight, even a partial de-rating of that complex would drag the index materially lower. This is the "painful repricing" Slok has warned about: a mismatch between priced expectations and actual ROI timelines resolving through price.
- Upside case — the S&P 493 finally answers. Cheaper, commoditized models let mainstream companies skip the expensive integration layer and embed intelligence directly into workflows. Revenue per worker and margins for the 493 inflect within the next two to four quarters, TFP accelerates, and the earnings base broadens enough to support current multiples with less reliance on the Mag 7. In this world, Slok's warning becomes a footnote to a delayed but genuine productivity boom.
By time horizon, the picture splits cleanly. In the short term, sentiment and liquidity dominate, and the Mag 7's 34.6% index weight means the benchmark can rise even while breadth improves among the 493. Over the medium term, the earnings cycle is the constraint: capex growth decelerates, depreciation rises, and the circularity between spending and reported profit becomes the central variable. Over the long term, the structural question is whether AI is a calculator that stays in the technology sector's profit-and-loss statement or a general-purpose technology that eventually rewires the rest of the economy. Slok's answer is that the story will be decided in the S&P 493, not the Mag 7. So far, the 493 has not answered — and the market is pricing as if it already has.
The market has priced AI as a profit shock for corporate America. The data so far say it is a profit shock only for the companies selling the shovels — and a cost center for everyone else.
Explore more exclusive insights at nextfin.ai.
