Apollo's economist says AI has delivered no payoff outside the tech giants
The short answer
Torsten Slok, chief economist at Apollo, published a note on 24 October titled No Signs of AI Payoff Outside the Tech Sector. His argument: revenue per worker and margins are rising for the largest technology companies and flat for the rest of the S&P 500, so the productivity story has not reached the firms doing the buying.
What the note says
The claim is narrow, and that is what makes it worth reading. Slok is not saying AI does nothing. He is saying the measurable gain shows up in the companies selling the infrastructure, not in the companies buying it. Margins and revenue per employee have moved for the largest tech names. Across the rest of the index, they have not.
His explanation is the boring one that keeps being right: general purpose technologies take years to show up in productivity statistics, because the gain requires reorganising the work, not just buying the tool. Electricity took decades. Computers took longer than the spreadsheet.
Why it matters for anyone buying AI tools
If the payoff is not showing up, it is usually because the tool was added on top of an unchanged process. A team that buys a chatbot and keeps the same approval chain gets a faster draft and the same bottleneck.
The test that matters is not adoption. It is whether one number moved: cycle time, cost per case, revenue per person, or hours per week on a specific task. If no number moved, the tool is a subscription, not an improvement.
Where the counter-argument sits
The obvious reply is that the measurement lags. Firms are still in year one of deployment, and the gains show up when the process is rebuilt around the tool, which takes two to three years. That is also what the historical record says, and it is why the note is a warning about expectations rather than a verdict on the technology.
The second reply is selection. The companies seeing gains are the ones with the data, the engineering staff, and the willingness to change how work is done. Everyone else bought the same model and got the same demo.
What to do with this
Pick one process this quarter, measure it before, change the process rather than adding another subscription, and measure it again. If the number does not move, stop paying for it. That is a cheaper strategy than waiting for the macro data to agree with the pitch deck.
Questions people ask
Does this mean AI is not working?
No. It means the measured gains are concentrated in the companies selling AI infrastructure, and have not yet appeared across the wider index.
Who is Torsten Slok?
Chief economist at Apollo Global Management. The note was published on 24 October 2026 and reported by TechCrunch the next day.
How long do these transitions usually take?
Historical general purpose technologies took one to three decades to show in productivity data, because the gain required reorganising work rather than buying equipment.
Read next: what an AI task actually costs and why your AI bill rises while prices fall.