Goldman Sachs says the AI growth boost is nearing its peak, while Apollo finds zero payoff outside tech

For more than a year, the AI spending spree has been one of the main forces holding up US economic growth. Now two of Wall Street’s most closely watched economists are suggesting that support is running out of road.

Goldman Sachs chief economist Jan Hatzius says the boost from AI investment is about to crest. Apollo chief economist Torsten Slok adds a harsher point: the productivity payoff that was supposed to take over hasn’t arrived for most companies.

Two economists, one uncomfortable picture

Start with Hatzius. His view is that AI-related capital expenditure, the money companies pour into chips, data centers, and related hardware, cannot keep climbing forever.

Goldman estimates US AI-related capex recently ran at roughly $600 billion. Hatzius expects that spending may still rise further before it eventually levels off.

He compares the expected plateau to earlier technology waves such as the railroads and electrification. Both transformed the economy, and both had a frantic construction phase that eventually gave way to something calmer.

The more awkward number is AI’s measured contribution to growth. Hatzius pointed out that AI added only 0.1 percentage points to US GDP growth recently.

Part of the reason is that much of the hardware is imported. When a US company buys foreign-made equipment, a big chunk of that spending leaks out of the domestic GDP figures. The economy gets the data center, but another country books a lot of the manufacturing.

Hatzius stressed that the move toward a fully realized AI economy should be a process measured in decades, not a burst that arrives in the next few quarters.

Then there’s Slok. In a note published Saturday, the Apollo economist went straight at the corporate return question.

“Despite massive investments and widespread hype around artificial intelligence, the technology has delivered zero financial payoff to corporate America outside of tech,” Slok wrote.

The Magnificent 7 versus everyone else

The core of Slok’s argument is a split between the biggest tech firms and the rest of the market.

On one side sit the so-called Magnificent 7: Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, and Tesla. On the other is what analysts call the S&P 493, meaning the remaining companies in the benchmark index.

According to Slok, revenue per worker and profit margins have risen significantly for the Magnificent 7. For the S&P 493, those same metrics have stayed flat.

Slok’s analysis also suggests mainstream corporate America increasingly treats spending on AI tokens, the metered units companies pay for when they use AI models, as a drag on margins. In accounting terms, AI is behaving more like a cost center than an investment in future growth.

How the AI growth story got here

Goldman has previously modeled AI as potentially adding up to a 1.5 percentage point boost to long-term US growth. That is a big number by macroeconomic standards, and it helped fuel the enthusiasm behind the spending wave.

What Hatzius and Slok describe is a gap between those two engines. The first may be nearing its peak. The second hasn’t meaningfully started outside the tech sector.

The broader debate over AI infrastructure has shifted as a result. The focus has moved toward import leakage, which shrinks the domestic GDP impact, and toward the time lag between spending money and seeing productivity rise.

What this means for markets and the economy

For investors, the most immediate takeaway is concentration risk. If AI’s gains are pooling in a handful of mega-cap companies, then the benefits of the AI trade are far narrower than the headlines suggest.

The research also raises a question about what happens when capex stops accelerating. If AI spending has been propping up growth, then a plateau removes a source of support, even if spending stays high. Growth figures respond to change, not just to levels. A flat line of spending, however large, contributes less to growth than a rising one.

There is also a policy dimension. A growth contribution of 0.1 percentage points is modest for an investment wave of this size. If the hardware keeps coming from abroad, more of the economic benefit of the buildout will land outside the US.

What to watch next is whether the S&P 493 starts showing movement in revenue per worker and profit margins. That would be the first sign that AI is moving from expense to engine for ordinary businesses. Also worth tracking: whether AI capex keeps climbing from the roughly $600 billion level Goldman estimates, or begins to flatten as Hatzius expects.