Goldman and Apollo Economists Warn AI Spending Outruns Payoff

Two of the market's most closely watched economists published back-to-back warnings within 48 hours that AI spending is outpacing its economic payoff — and both work at firms financing the buildout. Neither doubts where AI is headed, only how quickly the rest of the economy will get there.

At Goldman Sachs, chief economist Jan Hatzius estimates the growth impulse from the AI buildout will peak in late 2026 or early 2027. The handoff from historic capital spending to economy-wide productivity gains is hitting an air pocket: AI capex growth as a share of GDP is slowing, the boost from last year's tax depreciation incentives is fading, and the stock-market wealth effect is cooling. Goldman still expects a long-term payoff, but near-term labor-market disruption will weigh on growth before lifting it.

What does the data say outside Big Tech?

The productivity cushion is missing. Days before the Goldman note, Apollo chief economist Torsten Slok showed that profit margins and revenue per worker are rising among the Magnificent Seven but remain flat across the remaining 493 companies in the S&P 500. Mainstream companies now view token spending "not as a source of growth but as a cost center that weighs on margins." As Slok puts it, the fate of the AI boom "will be decided in the S&P 493, not the Mag 7" — and right now the 493 are barely moving.

In the boardroom, the caution evaporates. Goldman CEO David Solomon said in January that "the pace of capital investment is going to continue." Apollo chief Marc Rowan has called the AI transition "certainly" the biggest technology cycle of his career, one in which "almost every job will be enhanced or replaced."

Jamie Dimon, in his April letter to JPMorgan shareholders, predicted AI adoption would unfold "far faster than prior technological transformations, like electricity or the internet." BlackRock's Larry Fink has asked what happens to everyone else "if AI does to white-collar workers what globalization did to blue-collar workers," and Rowan expects "some amount of political upheaval."

Where both camps are right

The CEOs are right about the destination; the economists are right about the distance. Short-term capital is financing a transition whose payoff runs on a multi-year horizon. Even Solomon has acknowledged that enterprise implementation may unfold more slowly than people expect — and behind closed doors, his peers agree. In polling of 200 chief executives at the Yale CEO Summit in June 2025, 40% said AI hype has led to overinvestment, and more than 40% admitted AI investments have been allocated inefficiently.

That candor matters as capital pivots from training models to inference. Deploying models at scale exposes customer relationships, proprietary data, and brand reputation to a technology that still fails at an unacceptable rate — and agentic systems compound the risk, since a model that is 95% accurate at each step loses reliability as errors accumulate across steps.

How are customers and workers responding?

Badly, by the survey evidence. In Morning Consult polling for the 2026 Yale CEO Summit, 82% of American adults said human customer-service representatives have become too hard to reach, and 44% said a company's use of AI in customer service makes them less likely to engage with the business at all. Nearly twice as many consumers said AI takes longer or fails to resolve their problems (51%) as said it solves them faster (26%). Among employed adults, only 26% said AI mostly improves their productivity; roughly a third reported added friction through extra processes, verification, and rework. Among daily AI users, that friction share jumped to 44%.

McKinsey's 2026 State of AI survey found only about 6% of companies report a significant bottom-line impact from AI — defined as at least 5% of EBIT — a figure unchanged from a year earlier. About one in five respondents said operating costs, including tokens, have constrained their AI use. Andreessen Horowitz reported that as of April, barely 2% of US households were paying for an AI service, a share the firm calls growing but "still tiny in the big scheme of things."

Is the market overbuilt?

It may be, relative to the payback window that data center financing assumes. Much of corporate AI spending funds exploratory pilots rather than commercial deployments with measurable gains — and a pullback to reassess would be healthy, not a failure. Investors will need longer horizons, CEOs will have to frame modest pilot results as progress, and policymakers must guard against stranded power commitments and idle land from stalled data center projects.

The AI revolution is real. The herd has simply run ahead of it, and markets have priced in a payoff that runs on a much longer clock.