The economics of artificial intelligence (AI) are more uncertain today than they were two years ago, even as the technology itself continues to improve at a remarkable pace, according to Goldman Sachs Research’s Head of Global Equity Research Jim Covello.

Speaking on the recent Goldman Sachs Exchanges podcast The AI Investment Boom: When Will It Pay Off? with hosts Alison Nathan and George Lee, Covello said the industry has “gotten further away from” demonstrating clear returns on AI investment over the past two years.

Covello framed the central issue as economic, not technological.

“Look, at some point you got to make money,” Covello said on the podcast. “You make investments in a business so that you can generate returns and make money. And we’ve gotten further away from that over the last couple years instead of closer to it.”

He added that this does not imply AI will fail economically, but that the timeline for returns remains unresolved.

Covello noted that despite earlier expectations that spending would moderate if stocks underperformed, hyperscalers have instead increased capital expenditures.

“They have underperformed because of the significant investment in capex and the negative impact on their free cash flow,” Covello said. “But instead of cutting the capex, they’ve actually raised the capex.”

He said this dynamic reflects what he described as competitive pressure across the AI value chain.

“There is a tremendous amount of FOMO at every level of the supply chain,” he said.

Covello said the clearest economic beneficiaries so far have been semiconductor companies supplying AI infrastructure.

“The semiconductor companies are thriving at the economic expense of everybody above them in the chain,” he said.

He added that this is a departure from prior technology cycles, where semiconductor performance typically aligned more directly with downstream customer profitability.

A central theme of the discussion was whether enterprises are generating measurable returns from AI deployment.

Covello said the key question is whether enterprises are making or saving money from AI implementation.

“If they do, this technology is going to fulfill its promise,” he said. “If we’re having the same debate two years from now and we’re still saying, ‘Well, it’s early,’ then we might have a challenge.”

He also pointed to operational constraints within enterprises, including data readiness.

“In a lot of cases actually, the data isn’t ready to be agented yet,” Covello said, referring to the deployment of AI agents on enterprise data systems.

Covello referenced third-party surveys showing a gap between C-suite expectations and frontline worker experience regarding AI productivity gains.

He said surveys “pretty much say the same thing,” indicating that line workers are not seeing the level of productivity improvement executives anticipated.

Despite skepticism around near-term economics, Covello said AI companies are among the fastest-growing in corporate history in terms of top-line growth.

He also reiterated that the current market environment has supported strong investment flows into the sector, even amid uncertainty about returns.

“The market’s giving a long leash, as well it should,” Covello said, adding that investors are effectively waiting for evidence that profits will eventually flow through the ecosystem.

Covello said outcomes over the next several years will likely depend on whether enterprises can demonstrate consistent returns from AI adoption.

At present, he said, the economics remain unresolved, with value disproportionately accruing to semiconductor providers while upstream profitability remains uncertain.