Apollo Says AI Has Delivered ‘Zero Payoff’ Outside Tech — Goldman Isn’t So Sure

Last updated: 28 October 2026. Figures in this piece are sourced to primary publications and verified news reporting, linked inline; figures self-reported by the companies or economists involved are flagged as such.

The 60-second version

• On 24 October, Apollo Global Management’s chief economist Torsten Slok published a note arguing AI spending has produced “zero financial payoff” for corporate America outside the technology sector.
• Two days later, Goldman Sachs chief economist Jan Hatzius told clients the AI boom’s lift to US GDP growth is set to peak “in late 2026 or early 2027” — a more measured but directionally similar signal.
• Slok’s case rests on flat profit margins and no productivity lift across the “S&P 493” (the index minus the Magnificent Seven), even as those same companies ramp up AI and “token” spending.
• Hatzius doesn’t dispute the lag — he still expects AI to eventually lift productivity and profits economy-wide, but flags AI-linked borrowing as a drag on growth in the meantime.
• Neither economist is calling AI worthless. The disagreement is about timing, measurement, and how much slack the market has left to wait for the payoff to show up.

Key numbers from the two notes

What actually happened

Two of Wall Street’s most closely watched economists put out notes four days apart that, read together, amount to the most pointed mainstream pushback yet on the idea that 2026’s AI capital spending is already paying for itself.

Slok’s note, titled “No Signs of AI Payoff Outside the Tech Sector” and published through Apollo’s Daily Spark research letter, argued that despite “massive investments and widespread hype,” AI “has delivered zero financial payoff to corporate America outside of tech,” as first reported by Fortune. His evidence is aggregate: revenue per worker and profit margins are rising for the Magnificent Seven — the handful of companies selling AI infrastructure, chips and models — and essentially flat for the “S&P 493,” his own shorthand for everyone else in the index. He also cites the San Francisco Fed’s utilization-adjusted productivity series, which shows no detectable AI-driven step change.

Hatzius’s note landed two days later and struck a different tone, even while pointing the same way. Goldman’s base case, per Fortune, is still that AI investment “will ultimately pay off in terms of higher productivity and economywide profits.” But Hatzius flagged that the growth impulse from the AI boom — the extra GDP growth that capex and the associated “wealth effect” on consumer spending have generated — was already on track to crest “in late 2026 or early 2027,” independent of recent nerves about sustainability. He also pointed to a less-discussed channel: AI-linked borrowing has helped push developed-market long-term rates up roughly 100 basis points since early 2026, which he estimates could cumulatively shave 0.6 percentage points off GDP if it persists.

Concerns about AI capex outrunning AI revenue have been building all year, including in OpenAI’s most recent funding round, which valued the company at the same number as its compute bill. What’s new is that two economists at systemically important institutions are now making a related argument in the same week, using accounting and productivity data rather than market sentiment.

The data behind “zero financial payoff” — and its limits

Slok’s argument is narrower than the headline phrase suggests. He isn’t claiming AI tools don’t work, or that no company has found a use for them. He’s making an aggregate, index-level claim: Magnificent Seven profit margins have been climbing since the third quarter of 2023 — to roughly 30% — while margins for the “S&P 493” have stayed flat, in the 9%-to-10% range over the same period. Slok’s note doesn’t break the “S&P 493” figure down by sector, so it doesn’t show which industries are dragging on that average or by how much — a gap worth naming, since the aggregate could reflect weakness concentrated in a few sectors or spread evenly across all of them.

The glaring asymmetry is the “sellers” versus “buyers” framing: AI infrastructure and model providers are seeing margins and revenue per worker climb, while the companies buying their products show nothing. That matches the picture at the lab level, too. Anthropic’s own IPO filing disclosed an $80.7 billion net loss for the first half of 2026 even as revenue grew more than tenfold year-over-year, and deals like its $35 billion compute arrangement with Lambda show how much capital is pouring into the supply side. Money is visibly moving from buyers to sellers; Slok’s note adds evidence it isn’t yet moving back in the form of higher profits for the buyers.

There’s a methodological gap worth naming plainly: a flat aggregate margin doesn’t prove AI spending has no effect at the level of an individual firm or workflow. Many non-tech companies spent much of 2025 and early 2026 on pilots, data cleanup and integration rather than production deployments — costs that hit the P&L well before any offsetting gains do. Tariffs, wage growth and higher rates are also moving through the same margin lines Slok is reading, and his note doesn’t isolate AI’s effect from those. “Zero” is a strong word for what is, more precisely, a claim that the net effect isn’t visible yet in aggregate — a different and more defensible claim than “AI doesn’t work.”

Why the coverage is overselling the disagreement

Most headlines this week flattened two distinct arguments into one soundbite. Slok’s note is about a current absence of aggregate financial payoff. Hatzius’s is about the trajectory of a macroeconomic growth impulse — a narrower claim about the rate of change in GDP contribution, not a verdict on whether AI investment is worthwhile. Treating “Goldman says AI is peaking” and “Apollo says AI has zero payoff” as the same claim obscures that Goldman’s base case remains constructive on AI’s eventual payoff, while Slok’s is openly sceptical about the near term.

What most coverage has underplayed is the circularity risk this data points to. The Magnificent Seven’s rising margins are funded substantially by capital that those same companies, and the frontier labs they back, spend on each other’s infrastructure — chip and compute arrangements of the kind behind Qualcomm’s AWS deal, itself a stock warrant rather than a confirmed chip order when first announced. If the “S&P 493” genuinely isn’t buying enough AI-driven productivity to justify sustained token spending — Slok separately argues mainstream firms are starting to treat AI spend “not as a source of growth but as a cost center” — the revenue propping up the sellers’ side of this trade depends on a customer base that, by Apollo’s own numbers, isn’t yet seeing a return. That’s the tension Goldman’s rate-channel point gestures at, and it’s exactly the kind of structural signal worth tracking if you’re trying to spot an AI bubble early rather than after it has already popped.

Practical takeaways for builders, publishers and software teams

For anyone running AI-dependent products or workflows rather than trading AI stocks, the useful signal here isn’t “AI doesn’t pay off” — it’s that aggregate, company-wide measurement is the wrong level to look for a payoff at all.

Measure at the workflow level, not the P&L level. Slok’s data is index-wide; it says nothing about whether your specific use of AI in, say, customer support triage or content tagging is working. If you haven’t instrumented a before/after comparison for a specific workflow — time saved, error rate, cost per outcome — you don’t actually know your own answer to the question he’s asking about the economy as a whole.

Treat token spend as a cost line until proven otherwise. Budget it, track its marginal return, and be ready to defend the line item next cycle — Goldman’s own data shows some firms are already cutting AI budgets after the initial rush.

Watch the labour-market correlation. Goldman found heavier AI adoption correlates with weaker employment growth in non-manufacturing firms. If your AI deployment is substituting for hiring, that efficiency gain should show up in your own unit economics even if it never shows up in an index-level margin chart.

Don’t confuse capital-markets enthusiasm with product-market fit. The sellers’ side of this trade has every incentive to keep the narrative buoyant regardless of what buyers get back. Anchor build decisions to your own measured outcomes, not the state of the AI trade this week.

What we still don’t know

Apollo has not published the full methodology or underlying dataset behind its margin and productivity charts beyond citing Bloomberg, Macrobond and the San Francisco Fed, so independent replication of the exact “S&P 493” calculation isn’t currently possible from the public note alone. It’s also unclear how much of the flat aggregate margin reflects genuine AI underperformance versus a normal adoption lag — Slok’s note doesn’t separate pilot-phase spending from production deployment, which would meaningfully change the interpretation. Goldman’s note, as reported, doesn’t specify how much of the “AI-linked borrowing” behind its rate estimate is corporate versus sovereign, or how sensitive its 0.6%-of-GDP drag estimate is to further rate moves. And neither bank has said what specific threshold would cause it to revise its current view, which makes it hard to know what would actually falsify either economist’s position.

FAQ

Does “zero financial payoff” mean AI tools don’t work?
No. Slok’s claim is about aggregate profit margins across the S&P 500 excluding the Magnificent Seven — it says nothing about whether specific AI applications are useful at the level of an individual company or workflow.

What is the “S&P 493”?
It’s Slok’s own shorthand for the S&P 500 minus the Magnificent Seven (the small group of mega-cap tech companies, including the main AI infrastructure and model providers, that have dominated the index’s recent gains).

Does this mean the AI investment boom is about to collapse?
Not according to either economist. Goldman’s base case remains that AI investment will eventually lift productivity and profits economy-wide; its note is about the growth impulse’s trajectory, not a collapse call. Slok’s note is more sceptical about the near term but doesn’t forecast a downturn.

Why are two chief economists raising this in the same week?
Both notes draw on overlapping concerns — AI capex concentration, financing costs and the gap between spending and measurable returns — that have built through 2026 as AI infrastructure spending scaled into the hundreds of billions of dollars industry-wide.

Sources

• Torsten Slok, “No Signs of AI Payoff Outside the Tech Sector,” Apollo Daily Spark, 24 October 2026 (primary source)
• Fortune: “AI’s boost to the U.S. economy is nearing its peak, Goldman says, as Apollo finds ‘zero financial payoff’ outside Big Tech,” 26 October 2026
• explainx.ai: analysis of the Apollo note and its data sources, 24 October 2026
• Yahoo Finance (Bloomberg): “Apollo’s Slok Says AI Has Delivered Zero Payoff Outside of Tech,” 25 October 2026

Please follow and like us: