High interest rates drive up real cost of capital for AI
A report by consulting firm Gavekal, authored by analyst Will Denyer, claims that rising interest rates in the United States are the only real threat to the current cycle of investment in artificial intelligence infrastructure. After the Federal Reserve raised its benchmark rate on Wednesday (16), 10-year Treasury yields surpassed 5% for the first time since 2007, increasing the real cost of financing for data centers and other facilities in the sector.
Why a voluntary pause by labs is unlikely
Denyer argues that a coordinated slowdown is unlikely to occur, as any lab that slows down would quickly be overtaken by competitors. “If any leading AI lab slows down, competitors will close the gap, forcing it to speed up again or risk being overtaken,” he says. Competition, including with Chinese companies, and the pursuit of technological leadership make continued investment almost inevitable.
Impact of the Fed’s monetary tightening
The Fed’s interest rate hike not only raised nominal Treasury yields but also increased real interest rates, the type of cost that underpins much of the billion-dollar appetite for AI projects. The report cites the “Wicksellian spread,” the difference between the real return on invested capital and the real cost of capital. When this indicator falls below the historical median of the last 20 years, the risk of a widespread slowdown increases. Currently, the spread is close to but has not yet crossed this threshold. “Any further increase in real financing costs will raise the risk of an AI slowdown in a way that no blog post by Dario Amodei could ever provoke,” says Denyer.
Regulatory restrictions have limited impact
State moratoriums in New York and Texas, as well as other local restrictions, are already causing friction, but consulting firm SemiAnalysis estimates that they would only delay about 2.3 gigawatts of planned capacity in the United States. In contrast, the projection is for an additional 38 gigawatts in 2027, following 18 gigawatts in 2026 and 10 gigawatts in 2025. President Donald Trump has advocated for American leadership in AI, especially in competition with China, making tougher federal legislation unlikely in the short term.
Computing demand remains strong
So far, there are no signs of a slowdown in demand for computing capacity. Companies in the AI supply chain continue to report strong results, and the return on investment for the US corporate sector reached a record high in the last quarter. The shortage of processors, energy, and memory has been addressed through price increases, which stimulate supply and keep the investment cycle going.
Conclusion
The report suggests that while high interest rates pose a real risk, competitive and regulatory factors are unlikely to halt the current boom. Denyer notes that, in an extreme scenario, even fixed-income securities could serve as a hedge against a potential collapse in AI investments.