Torsten Slok has built a reputation for straight talk on markets and technology. As chief economist at Apollo Global Management, he now delivers a sobering message. AI adoption races ahead. Yet measurable productivity gains outside a narrow slice of tech firms remain modest at best. Markets, however, price in rapid returns. That gap sets the stage for trouble.
Slok laid out the risks in a note that Fortune reported on July 6, 2026. “The key issue is the length of the ROI runway outside the tech sector,” he said. “The bottom line is that a mismatch between current earnings expectations and the actual time firms need to generate ROI on AI investments could have significant implications for many AI company valuations today.”
His warning lands at a delicate moment. The Magnificent Seven lifted profit margins from 15 percent in the first quarter of 2023 to 25 percent in the first quarter of 2026. The rest of the S&P 493 hovered near 10 percent. Bloomberg’s broader 500 sat around 12 percent. Those figures come from Bloomberg and Macrobond data cited in the Fortune piece. The divergence tells its own story. Tech leaders reap rewards. Everyone else waits.
But wait they do. An MIT study found that only 5 percent of companies achieved meaningful return on investment from generative AI pilots. The other 95 percent saw little. Consultants and pilots abound. Real transformation proves harder. Regulatory hurdles, data protection rules, and the sheer difficulty of reshaping workflows slow everything down. So firms experiment. They optimize tokens. They chase metrics that look good in reports. Actual output gains stay elusive.
Slok’s January 2026 analysis painted a more optimistic picture. In “Quantifying the Productivity Gains from AI,” published via Apollo Academy, he and co-authors argued the economy sits in the early stages of another productivity boom. AI spreads faster than PCs or the internet did. Surveys showed 67 percent of firms reporting labor productivity increases greater than 5 percent in the prior year. Time savings from generative AI ranged from 1.4 percent to 1.8 percent of hours worked overall, reaching 3.5 percent in professional services.
Capital spending tells part of the tale too. Hyperscalers pour money into data centers at a pace that exceeds telecom investment during the dot-com years when measured against GDP. Projections point to shares near 2 percent. The earlier boom peaked lower. Slok sees parallels to the 1990s IT surge. Creative destruction and sectoral reallocation could drive total factor productivity gains of 0.2 percent to 0.5 percent at peak, with cumulative effects of 2 percent to 6 percent over a decade or more.
Yet his recent comments strike a cautionary note. In a June 30 piece covered by Yahoo Finance, Slok stressed that payoffs “could still be years away.” Markets assume a hockey-stick trajectory measured in months. Reality may stretch to five years. “Equity markets priced for instant earnings growth will face a painful repricing if the productivity hockey-stick takes five years rather than five months,” he wrote. Companies will cut AI budgets without quick wins. Spending slows. Valuations adjust. The process hurts.
History offers a map. The Solow Paradox, named for economist Robert Solow’s 1987 observation that “you can see the computer age everywhere but in the productivity statistics,” haunted the 1970s through early 1990s. Massive IT spending produced no clear efficiency jump at first. Then the late 1990s delivered acceleration. Labor productivity rose. Total factor productivity followed with a lag. A May 2026 Federal Reserve Bank of San Francisco research brief, referenced in Fortune’s May 27 coverage, draws the parallel. “If today mirrors what we experienced in the mid-1990s, we may be in the early stages of a productivity boom driven by AI that will only become clear in retrospect,” the researchers wrote.
Workers already move faster. A London School of Economics study found AI delivers the equivalent of one extra workday per week in time savings for some. Yet that speed often translates into more tasks rather than fewer hours or higher efficiency at the organizational level. Harvard Business Review cases document the pattern. Employees redirect saved time into additional work. Burnout risks rise. Cognitive load increases. Economy-wide efficiency metrics stay stubbornly flat.
Real-world experiments reveal the friction. Ford hired 350 experienced engineers, the so-called “gray beards,” to oversee AI-driven automation across 33 plants equipped with more than 1,000 cameras. IBM cut thousands of roles while tripling entry-level hiring to reshape its workforce around new tools. Ricoh spent $500,000 on consultants and $200,000 monthly on AI services only to find the automated approach cost three times more than manual processing. Headcount dropped modestly from 44 to 39. These examples, drawn from the July 6 Fortune article and earlier reporting, show promise mixed with expense and limited scale.
Slok remains no doomsayer. He has long argued AI will create more jobs than it destroys, invoking the Jevons paradox. Cheaper, more capable tools spur greater demand. Small business entrepreneurship could flourish. An industrial renaissance in semiconductors, pharmaceuticals, and defense might follow. His May 2026 Apollo presentation on the impact of AI on the economy and financial markets highlighted those upside possibilities alongside datacenter and energy spending trends.
Still, the near-term tension dominates boardroom conversations. A National Bureau of Economic Research survey of nearly 750 executives, released in March 2026 and covered by PSCA, found widespread AI investment. Productivity gains registered as modest but rising, strongest in high-skill services and finance at about 0.8 percent implied growth. Lower-skill sectors saw roughly 0.4 percent. Executives expect acceleration this year. Whether that materializes fast enough to justify current stock multiples is the open question.
Recent market chatter echoes the debate. On X, analysts and investors traded views throughout early July 2026. Some pointed to $650 billion in annual AI capex and asked whether 7 percent annual productivity growth would justify it. Others warned of “tokenmaxxing” as a sign that implementation hits walls. Chinese hedge funds flagged a potential “super bubble.” BlackRock noted that the bubble question hinges on whether AI converts scarcity into abundance.
Schroders’ 2026 outlook asked if the moment resembles a productive bubble like railways or the internet, where infrastructure spending eventually pays off, or something less sustainable. Gavekal’s distinction between unproductive speculation and productive build-out tilts many observers toward the latter. Yet even productive bubbles can inflict pain during adjustment.
Slok’s core point cuts through the noise. The divergence between front-loaded valuations and slower cash-flow realities cannot last forever. Firms outside tech face longer integration periods. They encounter regulatory friction and organizational inertia. If earnings fail to match elevated expectations, spending plans shrink. Multiples compress. The repricing begins.
And markets hate surprises. They especially dislike when the surprise confirms what some economists have suspected all along. The productivity statistics have not yet caught up to the hype. Workers feel faster. Balance sheets show higher costs in many cases. Aggregate efficiency metrics lag. The 1990s taught patience. This decade may demand it too.
Slok’s analysis does not dismiss AI’s potential. Far from it. His own research forecasts meaningful cumulative gains over 10 to 20 years under various diffusion scenarios. He simply insists on realism about timing. Five months or five years makes all the difference for valuations built on instant transformation.
Investors now weigh the evidence. Corporate pilots proliferate. A handful deliver clear wins. Most deliver lessons. Hyperscalers keep spending. Smaller firms watch closely and hesitate. The data center buildout continues. Energy demand climbs. Job shifts occur. Yet the broad productivity surge that would ratify today’s prices stays just out of reach.
So the debate simmers. Is this the early stage of a boom that history will later recognize? Or does the gap between expectation and delivery widen until something breaks? Slok leans toward the former in the long run. He warns of the latter in the short term. Painful repricing remains a live risk.
Executives at Ford, IBM, and Ricoh navigate that tension daily. So do portfolio managers who own the Magnificent Seven at premium multiples. The next several quarters of earnings reports will prove critical, as Slok told CNBC in early July. Markets will parse every productivity comment, every capex update, every ROI disclosure.
The technology advances. Integration takes time. History suggests the gains arrive eventually. The question is whether markets can endure the wait without a sharp reset. Slok thinks the answer may be no. His caution deserves attention.
WebProNews is an iEntry Publication