One of Wall Street's most widely read economists has looked for the AI dividend in ordinary American companies and found nothing yet. Torsten Slok, chief economist at the private equity firm Apollo Global Management, set out the case in his Daily Spark note of 24 October, and TechCrunch picked it up a day later. It matters because almost every pitch for AI software, from chat assistants to coding agents, rests on the promise that it will lift productivity and profits for the companies that buy it.
Slok splits the S&P 500 in two. On one side are the seven giants: Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla. On the other are the remaining 493 companies, which he treats as a stand-in for mainstream corporate America. His charts show revenue per worker, adjusted for inflation to the end of 2019, rising for the seven and flat for the rest, and profit margins following the same split. A third chart, built on a San Francisco Fed measure of total factor productivity, shows no lift in productivity growth across the economy.
His reading of why goes further than the numbers. Start-ups backed by venture capital are spending more and more on tokens, he writes, but most large firms do not need the newest model and will not pay for it, and many now see token spending as a cost that eats into margins. As long as heavy use of AI agents stays confined to a small corner of the economy, he argues, the effect on growth will be small. He also doubts that cheaper models will change that, because business decisions turn on judgement, accountability and trust, which he says AI does not understand.
The finding sits awkwardly beside the steady flow of vendor case studies that report hours saved per employee. Both can be true. Time saved inside a team does not reach the income statement unless a company changes what it sells, how many people it hires or how fast it grows, and Slok's charts measure only the end of that chain.
The note has clear limits. It is a short market commentary built on a handful of charts, not a peer-reviewed study, and it does not test causes. Margins and revenue per worker across 493 very different firms are coarse measures that can hide gains in some sectors and losses in others, and they say nothing about costs avoided. Apollo also attaches its usual disclaimer that the views are the speaker's opinion. What the note does well is set a test: if AI is changing ordinary businesses, these lines should start to move, and that is worth checking again in a year.