Warsh says rise of AI will prove ‘very bumpy,’ trigger demand shock

A demand shock has jarred the U.S. economy for several months as large technology companies — known as hyperscalers — rush to secure the electricity capacity, cooling systems, server hardware, advanced semiconductors and other capital goods essential to the AI build-out.

U.S. investment in AI from 2025 through 2032 will total $10.3 trillion, or an average of 3.63% of gross domestic product per year, according to Stijn Van Nieuwerburgh, a Columbia University professor.

“Investment is flowing into AI on a scale that will likely exceed the vast sums that went into building the railroads, the electricity grid and the internet,” IMF Managing Director Kristalina Georgieva said Thursday.

Soaring investment has pushed up demand for capital from the largest tech companies, which for years have had comparatively limited borrowing needs.

Although the hyperscalers “have profitable core businesses, they are increasingly turning to debt markets for financing,” Georgieva said in a blog.

“Just in the last two years, their share of investment-grade bond issuance has increased more than tenfold — from less than 1% to 10% and growing,” she said.

Since mid-September borrowing by the hyperscalers helped push up the yield on the benchmark 10-year treasury note beyond 5%, increasing the cost of borrowing for consumers and businesses.

The hyperscalers face risks from their colossal borrowing, according to Gartner.

More than $3.5 trillion of AI infrastructure spending may be exposed to impairment by the end of 2030 if providers of the technology sustain their spending at current levels, Gartner said Wednesday in a report.

“Our research indicates that enterprise and consumer AI revenue will be insufficient to support profitable returns on all planned AI capital expenditures” between 2027 and 2030, Gartner Senior Director Analyst Will Sommer said in a statement.

Huge investment in AI also may jolt the financial system and broader economy, Georgieva said.

“The risk is that expectations may run rapidly ahead of productivity gains, setting the stage for near-term disappointment,” she said, noting that under Amara’s Law “we tend to overestimate the effect of a new technology in the short run and underestimate its effect in the long run.”

“Big revenue or earnings disappointments by AI developers could undermine expectations of future profitability, dealing a hit to equity prices,” she said.

“In a severe scenario, we could see a market correction that cascades through the highly leveraged and circular AI ecosystem,” Georgieva said. “Given the sizable cross-border holdings of AI assets, the losses could spread around the world.”

Warsh expressed optimism that, despite the hazards, AI will prove a boon to the global economy.

“It’s an exciting development,” he said, adding “I think a decade from now they’ll look back and say, ‘This was hugely consequential.’”