▲ A data center in Georgia, USA (File Photo)
Cracks are appearing in the estimated 10 trillion-dollar U.S. artificial intelligence (AI) infrastructure investment boom, the Financial Times (FT) reported on the 18th local time.
As the AI boom enters its fourth year, data center construction plans are running into practical hurdles, the report explained.
Citing data from regulatory intelligence firm RegLayers, the newspaper reported that at least 47 major data center projects have been delayed, withdrawn, or blocked from approval so far this year.
RegLayers noted that this figure is a "conservative estimate" focused solely on the largest projects.
Earlier this month, investment bank Bernstein also estimated that only about 35% of planned data center projects in the United States are viable.
The Financial Times added that its own investigation found that over 1,000 local communities across the U.S. have attempted to introduce moratoriums or other restrictive measures against AI infrastructure development this year.
Financial conditions are also worsening.
Rising interest rates and a deteriorating bond market have made it difficult for neocloud companies—which lease and operate AI data centers—to secure financing.
CoreWeave shares plunged 23.6% last week, and the trading price of one of the corporate bonds issued by the company is in distressed debt territory.
SoftBank-backed SB Energy and Australia's Firmus Grid, backed by Nvidia, have delayed or suspended their listing plans.
Multiple sites being developed by Oracle for "Stargate"—the mega-scale data center project by OpenAI, Oracle, and SoftBank—are facing schedule delays, and Anthropic is increasingly likely to cancel a 45-billion-dollar contract with British neocloud firm Enscale.
The newspaper pointed out that AI companies have prioritized speed to avoid computing power shortages, even if it means relying on inexperienced partners.
Some in the industry have compared the situation to the "Wild West," it said.
With capital flooding into startups and due diligence for new projects rushed, AI companies are even absorbing short-term cost increases.
Stijn Van Nieuwerburgh, a professor at Columbia Business School, estimated in a Brookings Institution paper that U.S. AI infrastructure spending between 2025 and 2032 will reach 10.3 trillion dollars, accounting for 3.6% of annual gross domestic product (GDP).
This is about 1.5 times the peak of the 19th-century railroad boom and more than three times that of the internet boom.
He warned that massive liabilities are "all hidden off the balance sheet," adding that even a 10% drop in cash flow could wipe out equity value.
In this regard, tech media outlet The Information reported, citing executives in the data center and neocloud industries, that the current hundreds of neoclouds will likely consolidate down to about 7 to 10 within a few years.
Two individuals offered this outlook, and multiple industry voices expressed that consolidation is inevitable, the outlet reported.
The publication noted that even if a company that leases data centers like CoreWeave collapses, the impact on construction itself would be minor because the facilities are easy to find new tenants for; however, if companies that build their own facilities like Nebius, Iris, and Enscale falter, it would pose a heavier burden on Nvidia.
David Driggers, CEO of early-stage neocloud Cirrascale, predicted that "consolidation will definitely happen," adding that asset management firms or telecom and fiber-optic companies could step in for acquisitions.
(Photo: AP, Yonhap News)
※
Copying, redistribution, and unauthorized use in AI training are strictly prohibited.