Increased spending on artificial intelligence and the rising electricity cost of building data centres are creating new inflation pressures, while investors are still waiting for long-promised productivity gains.
While AI’s rapid rise initially sparked fears of major job losses, the technology, still in its infancy, is contributing to higher prices. Few examples of technology have promised so much and affected so many areas, while still in their early phase.
“The bottom line is that the AI spending boom is stoking both employment and inflation,” Apollo chief economist Torsten Slok wrote in a recent blog post.
At the heart of the AI boom is the surge in data centres, which house servers, networking equipment and the support systems necessary to power tools such as ChatGPT and cloud services.
There are more than 3,000 operational data centres in the US and about 1,500 more under construction or in early development, a Pew Research Centre report said in April.
More than $5 trillion is estimated to build out the AI ecosystem by 2030. And the four hyperscalers – Alphabet, Amazon, Meta and Microsoft – are driving a massive investment surge, with cumulative AI-related capex for this year projected at $650 billion, a 67 per cent increase from 2025.
AI is seated within the nexus of US investment relations with the Gulf, which has backed projects, including the $500 billion Stargate Project in Abilene, Texas.
However, these data centres require vast amounts of energy. Rystad Energy estimates the US has more than 100 gigawatts of data centre demand between 2024 and 2035. The Stargate Project alone is expected to require 7GW of power, compared to the 6GW needed to run New York City.
In its annual energy outlook report, the US Energy Information Administration said that after 15 years of relatively flat consumption, demand has risen by 2.1 per cent on average over the past five years largely because of data centres.
Jay Zagorsky, a professor at the Questrom School of Business at Boston University, said the AI boom is leading to higher prices on chip supplies, concrete and steel, and higher wages for construction workers building the facilities.
“All these strains are boosting costs and increasing inflation in a few key sectors of the economy,” he said.
Policy debate intensifies
The question facing policymakers today is how much AI will drive inflation, when it will start boosting productivity, how it will reshape the labour market and whether it will eventually ease price pressures.
“The biggest uncertainty for the Fed is whether this is another transitory effect or a new super cycle for memory prices,” Michael Pearce, chief US economist at Oxford Economics, told The National. “The chip cycle has historically been very volatile, and prices could drop back quickly once supply adjusts and the bottleneck eases.”
The view from the White House and new Federal Reserve chair Kevin Warsh is that AI will be a “significant disinflationary force” that could pave the way for lower interest rates.
Meanwhile, Fed officials are warning the promised productivity gains from AI could arrive after the nascent technology's inflationary bump, with officials growing more concerned that inflation, the albatross that continues to hang around the central bank's neck, remains above their 2 per cent target.
economist
San Francisco Fed President Mary Daly on Thursdasy said she believes AI's impact on inflation is "not a pressing issue" for monetary policy, saying that while it could lower inflation in the next five to 10 years, its long-term effects sit outside the timerame that guides Fed policy decisions.
She also said during the event she does not believe AI is behind the recent surge in inflation, which has been driven by tariff and higher energy costs.
Mr Zagorsky said that even if AI productivity gains do arrive, their impact will be limited to only a handful of sectors and will not lead to lower inflation.
“I don't see how AI will dramatically lower energy costs, which currently are a key driver of inflation,” he said. “If anything, AI will likely boost energy costs since running data centres is energy intensive.”
Bond market in check
AI enthusiasm's impact on the stock market is well documented, with Wall Street's major indexes reaching all-time highs in recent weeks despite uncertainty over the Iran war. However, a question is beginning to unfold in the bond market, where AI debt-funded capex is adding to supply pressures.
A Cushman & Wakefield analysis this month found the major hyperscalers and AI infrastructure borrowers issued a combined $121 billion in US corporate bonds last year, well above the 2020-24 average of $28 billion.
While the yield on the 10-year Treasury has risen in recent months, a Pimco analysis said the recent rise in longer-dated US Treasury yields is a larger reflection of the uncertainty surrounding the Iran war and the Federal Reserve's expected rate path.
It added that while the yield on the benchmark has risen by about 51 basis points since the Iran war began, only 13 basis points reflects a higher-term premium. Lotfi Karoui, multi-asset credit strategist at Pimco, wrote that while debt-funded AI capex could become a driver of risk premium, the transition may take place over years.
“The idea that the year-to-date surge in AI-related debt issuance has been a contributor to the recent rise in long-dated US Treasury yields appears overstated,” he wrote in an analysis.
The most recent test comes as Apollo and Blackstone are seeking to arrange a $36 billion debt deal related to Anthropic's AI infrastructure buildout, Bloomberg reported, with Broadcom backstopping the largest portion of the debt. Under the structure of the deal, the borrowed fund would buy customised Google chips called tensor-processing units, which Anthropic would lease without carrying it on its balance sheet.
Peter Andersen, founder of Andersen Capital Management, said that, for now, the high-yield market is not showing “any indigestion about this”.
“The bond market is still showing continued faith in this and nobody's blinking yet in terms of saying that we've overbuilt,” he said.
