Outside Big Tech, corporate America has spent heavily on AI and has almost nothing to show for it on the bottom line. That’s the conclusion of Torsten Slok, chief economist at private equity giant Apollo Global Management, in a note that TechCrunch AI reports is getting wide attention. Slok’s verdict is blunt: “Despite massive investments and widespread hype around artificial intelligence, the technology has delivered zero financial payoff to corporate America outside of tech.”
It’s a sharp claim from someone whose charts get passed around trading desks every day. It also lands at a time when boards are being asked to approve another year of AI budgets.
What Slok Actually Measured
Slok looked at what he calls the “S&P 493.” That’s the S&P 500 minus the seven giants driving most of the AI story: Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla.
For the remaining 493 companies, he found:
- Flat profit margins. No visible lift in profitability since the generative AI wave began.
- Flat revenue per worker. This is a rough stand-in for productivity, and it isn’t moving.
- No evidence of productivity gains from AI adoption at the aggregate level.
Take the Magnificent Seven out of the picture and the AI boom looks a lot more like an AI expense.
Lagging or Stuck?
The obvious counterargument is timing. Non-tech companies adopt technology slowly. They have to rework processes, retrain staff and clean up messy data before any gains show up. Economists have seen this before. In 1987, Robert Solow famously quipped that you could see the computer age everywhere except in the productivity statistics. The gains eventually arrived in the late 1990s, nearly a decade later.
Slok isn’t buying the “just wait” story, though. He argues it will be tough for companies to “turn increasingly cheap intelligence into higher revenue growth and higher profit margins.” His reasoning goes after what most businesses actually need from AI:
“AI may be a great calculator, but in business, intelligence is mostly judgment.”
He adds that “AI does not understand accountability, trust, culture, unwritten rules, ethical trade-offs or fairness,” which are the things that shape reputational risk and, in the end, revenue and margins.
What stands out to me is that he isn’t questioning whether AI works. He’s questioning whether cheap intelligence is what limits most businesses in the first place. If a company’s real constraint is judgment, relationships or distribution, a faster calculator doesn’t change much.
Why This Matters Now
Slok isn’t the only one saying this. As TechCrunch AI notes, recent studies have also found that AI hasn’t yet produced significant productivity gains. MIT’s widely cited 2025 research on enterprise generative AI pilots reached a similar conclusion: most projects showed no measurable return.
The timing matters for three reasons:
- The capex gap is getting wider. Hyperscalers keep raising AI infrastructure spending, and that bet only pays off if the S&P 493 eventually becomes paying customers at scale.
- Budget scrutiny is rising. CFOs who approved experimental AI spending in 2024 and 2025 now want proof, not pilots.
- Valuations assume broad adoption. If the gains stay locked inside tech, the market story behind AI stocks gets much harder to defend.
The Other Side
It’s worth pushing back a little. Aggregate numbers can hide real wins. A few strong deployments in customer service or software development won’t show up in index-level margins, especially when many companies are reinvesting their savings or passing them on through lower prices. Revenue per worker is also a blunt tool. It can’t easily separate AI’s effect from interest rates, labor markets or tariffs.
Still, “it’s in there somewhere” isn’t a satisfying answer after years of heavy spending.
What Businesses Should Do
Slok’s note gives you a useful filter for your own AI plans:
- Measure outcomes, not usage. Seat counts and prompts per day aren’t ROI. Tie every deployment to margin, revenue or cycle time.
- Aim at real bottlenecks. Put AI where cheap intelligence actually unlocks value, like document processing, code and support triage. Don’t expect it to replace judgment-heavy work.
- Keep humans accountable. Slok’s point about trust and accountability is a design requirement. Build workflows where people own the decisions.
- Set a deadline. Give pilots 6 to 12 months to show measurable gains, then scale them or cut them.
My take: the next year will split companies into two groups. One group redesigns how work gets done around AI. The other just bolts a chatbot onto old processes. Only the first group is likely to show up in Slok’s numbers. The full report is available at TechCrunch AI.