Apollo data: AI ROI is a Big Tech story, not yours

Apollo Global Management's chief economist has a blunt message for every CEO who approved an AI budget this year: the numbers aren't moving. Revenue per worker at S&P 493 companies — that's the S&P 500 minus Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, and Tesla — has sat flat at about $550,000 since 2019. Profit margins? Also flat. Meanwhile, those seven firms watched their revenue per employee jump from $550,000 in 2022 to almost $800,000 by 2026. If you're not in that group, the data says AI hasn't paid you back yet.

Apollo published its analysis last weekend, and SiliconANGLE and TechCrunch both covered it yesterday. Torsten Slok, identified by TechCrunch as Apollo's chief economist, drew his conclusions from graphs tracking three metrics across S&P 500 companies: revenue per worker, profit margins, and productivity growth. The Magnificent Seven's lines go up sharply from the third quarter of 2023 onward. The S&P 493's lines go nowhere.

Slok's verdict is direct: 'Despite massive investments and widespread hype around artificial intelligence, the technology has delivered zero financial payoff to corporate America outside of tech.' He argues the core problem isn't model quality or price — it's judgment. 'AI may be a great calculator, but in business, intelligence is mostly judgment,' he wrote. 'AI does not understand accountability, trust, culture, unwritten rules, ethical trade-offs or fairness, all of which shape reputational risk and ultimately drive revenue growth and profit margins.'

The shift already happening on the ground

Slok also notes a behavioral split forming right now. Venture-backed startups are still ramping up AI token spending. More established firms are pulling back. His read: 'Mainstream corporate America is pulling back and starting to view token spending not as a source of growth but as a cost center that weighs on margins.' And the kicker — 'Mainstream corporate America does not need the latest model and is unwilling to pay for it.'

That's a significant claim. It means the AI vendors' pitch — that cheaper, faster models will eventually unlock mass adoption — may be missing the point entirely. Slok isn't convinced cheaper will fix it: 'The question looking ahead is whether, even as models become commoditized, mainstream corporate America can turn increasingly cheap intelligence into higher revenue growth and higher profit margins.' His answer: 'That may be harder than it sounds.'

Why this matters more than the usual AI skepticism

Here's the thing — this isn't a think-tank op-ed or a journalist's hot take. Apollo Global Management manages over half a trillion dollars in assets and has deep visibility into corporate financials across industries. When their chief economist looks at S&P 493 margin and productivity data and sees nothing, that carries weight. It's not a survey of feelings about AI. It's financial outcomes.

The honest counterargument is timing. Transformative technologies often take years to show up in aggregate productivity numbers — electrification famously took decades to move the needle. AI adoption is still uneven, implementations are immature, and two years of data may simply be too short a window. Slok himself frames his conclusion as 'so far,' not 'never.' That's fair.

But 'it might work eventually' is not a business case. And the specific challenge Slok names — AI's inability to handle judgment, culture, and ethical trade-offs — isn't a problem that cheaper tokens solve. A model that costs half as much but still can't navigate a difficult client relationship or a reputational decision isn't twice as useful. It's still not useful for the hard stuff. The productivity gains showing up in the Magnificent Seven are largely concentrated in companies that build and sell AI infrastructure, not companies that buy and deploy it. That distinction matters enormously for how you think about your own spending.

Slok puts the stakes plainly: 'The bottom line is that the AI story will be decided in the S&P 493, not the Magnificent Seven, and so far there is no evidence that revenue growth or margins are improving outside of the tech sector.' If that doesn't change, the current AI investment cycle starts to look less like a productivity revolution and more like a very expensive experiment with unclear sponsors.

What to do with this on Monday morning

Three things worth acting on this week. First, pull your AI spend line and ask what it's attached to. Not 'what tools are we paying for' — what specific business outcome was this meant to move, and has it moved? If nobody on your team can answer that clearly, you don't have an investment, you have a subscription.

Second, if you're in a vendor conversation right now, ask them directly: which companies in your industry have seen measurable margin or revenue-per-employee improvement from this product, and can you talk to them? Vendors who can't answer that question with a real reference are asking you to fund their proof of concept.

Third, take Slok's judgment point seriously before your next deployment decision. The workflows where AI is most likely to show up in your margins are narrow, repetitive, and low-stakes — document processing, first-draft generation, routing. The ones that touch client relationships, reputation, or complex trade-offs are exactly where he says AI still falls short. Knowing which bucket your use case sits in is the difference between a cost that pays off and one that just weighs on margins.

Watch for whether S&P 493 productivity or margin data starts to move in 2027 earnings reports. If it doesn't, Slok's skepticism about cheaper models solving the problem will look more and more correct — and the vendors promising ROI will have a harder case to make. My bet is the gap narrows slowly, not dramatically, and that most mid-market leaders will still be waiting for the breakthrough quarter two years from now.