The AI Payoff Gap Leaving Most Companies Behind

The AI Payoff Gap is the uncomfortable split at the heart of this autumn's artificial intelligence debate: a handful of giant technology firms are turning heavy AI spending into rising sales per worker and stronger margins, while the rest of corporate America is, so far, getting little visible return at all. That is the argument in a new note from Apollo Global Management chief economist Torsten Slok, published on Saturday, and it lands just as large technology companies prepare to report earnings later this week. According to reporting on the note, Slok argues that the technology has delivered no measurable financial payoff outside the technology sector, even after massive investment and widespread hype.

Slok's method is a simple split of the broad market index into two groups. On one side sit seven giant technology firms: Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla. On the other side sit the remaining four hundred ninety-three companies in that index, spanning banks, retailers, manufacturers, airlines, drug makers and energy firms. Comparing those two groups over recent years, Slok finds that revenue per worker has climbed steeply for the giants while barely moving for everyone else. Profit margins tell a similar story, with the giants keeping roughly three dollars of profit from every ten dollars of sales, while the rest keep roughly one dollar from every ten. Productivity growth, the broadest measure of whether an economy is getting more output from the same inputs, shows no clear acceleration either, according to the note.

That framing matters because it reframes a familiar question. For two years, investors have asked whether AI spending will pay off at all. Slok's answer is that it already pays, but only for the companies selling the tools, building the models and renting out the computing power. The buyers, so far, look more like cost bearers than beneficiaries. In his words, mainstream firms are starting to treat token spending, the metered cost of running AI models, as a burden on margins rather than a source of growth. The note adds that as long as agentic AI and token spending stay confined to a small slice of the economy, the macroeconomic effects of the technology will stay limited.

What the spending concentration shows

One of the starkest charts in the note concerns who actually pays for AI at scale. According to a summary of the research, the top tenth of firms by AI spending account for nearly all model-serving and inference expenses, and a similarly overwhelming share of spending on rented AI computing capacity. Venture-backed firms dominate that top tier: roughly eight in ten of the heaviest AI spenders carry venture funding, compared with roughly one in eight among typical spenders. That concentration helps explain why the aggregate numbers look so lopsided. When a small group spends at a completely different scale from everyone else, averages for the whole market hide more than they reveal.

The price trend cuts both ways, and it is the second big piece of evidence in the AI Payoff Gap debate. Model providers have cut some inference prices by as much as four fifths in recent months, a collapse that analysts attribute to competition from cheaper rivals, better inference efficiency and the perishable nature of spare computing capacity, as tracked in a daily digest of the price moves. Cheaper intelligence should, in theory, help smaller buyers most, since cost was their biggest barrier. Slok is sceptical, arguing that it will still be hard for ordinary firms to turn increasingly cheap intelligence into higher revenue growth and stronger margins. In his framing, business intelligence is mostly judgment: accountability, trust, culture, unwritten rules, ethical trade-offs and fairness, none of which a model automatically supplies.

There is a counterpoint, and it comes from inside the technology industry itself. Executives at IBM have argued that open-weight models, whose underlying weights are freely available, can now handle routine work such as summarising documents and reviewing contracts at a fraction of the cost of frontier systems, making small on-premises deployments attractive for everyday tasks. If that claim holds, the payoff may simply be arriving through cheaper, narrower tools rather than through the flagship models that dominate headlines. The distinction matters for readers weighing a subscription or a pilot project: the most expensive model is not necessarily the one that changes a workflow, and a modest tool that fits an existing process may beat a powerful one that does not.

Why the next earnings season is a test

The timing of Slok's note gives it extra weight. Large technology firms report earnings this week, and their results will show whether AI revenue is still compounding for the sellers even as buyers pull back. If the giants report another quarter of rising sales per worker while the rest of the market reports flat margins, the AI Payoff Gap will look less like a temporary lag and more like a structural divide. If, instead, non-technology firms start reporting savings from AI in customer service, logistics or back-office work, the sceptics will have to revise their case. Either way, the split Slok describes gives readers a cleaner way to read the headlines: ask not whether AI works in general, but who is actually banking the gain.

For young workers, the stakes are practical rather than abstract. The firms most likely to hire, promote and train early-career staff are precisely the ordinary companies in the second group, not the seven giants. If those employers see AI mainly as a cost, they may slow hiring, freeze tool budgets or demand proof of return before approving new software. If they start seeing real savings, they may invest in training that makes AI fluency a genuine career asset rather than a line on a job posting. The lesson is not to ignore AI, but to judge it the way Slok judges it: by revenue, margins and productivity, not by demos. A tool earns its place when it changes one of those three numbers, and until it does, it is a cost to manage, not a miracle to celebrate. For more on how AI infrastructure is reshaping the wider economy, see our report on Starcloud's plan to put data centers in orbit, and for how ordinary users are actually adopting AI tools, read how Canadians are turning to AI for health answers. Creators weighing the economics of new platforms may also want how X's rewards change the way creators get paid.