47 U.S. Data Center Projects Hit Snags as Cracks Emerge in AI Infrastructure Boom

[Financial News] The UK's Financial Times (FT) reported on the 18th (local time) that cracks are beginning to appear in the U.S. artificial intelligence (AI) infrastructure investment boom, estimated at $10 trillion.
With the AI boom having lasted more than three years, data center construction plans are running into practical challenges.
Citing data from regulatory information firm RegLayer, the FT reported that at least 47 large data center projects have been delayed, withdrawn, or had their approvals blocked since the start of this year.
RegLayer said the figure was a “conservative estimate” covering only the largest projects.

Earlier this month, investment bank Bernstein also estimated that only about 35% of planned data center projects in the U.S. were viable.
The FT added that its own research found that more than 1,000 communities across the U.S. had tried this year to impose a moratorium (temporary pause) or other restrictions on AI infrastructure development.
Funding conditions are also worsening.
Rising interest rates and a deteriorating bond market are making it difficult for neocloud companies, which lease and operate AI data centers, to raise funds.
CoreWeave's stock fell 23.6% last week, and one of the corporate bonds it issued is trading in distressed-debt territory.
SB Energy, a SoftBank affiliate, and Australia's Firmus Grid, which has received investment from Nvidia, among others, have delayed or put on hold their plans to go public.

Several sites Oracle is developing for Stargate Project, the massive data center project involving OpenAI, Oracle and SoftBank, are also facing schedule delays, while the likelihood that Anthropic will terminate its $45 billion contract with UK neocloud Nscale is growing.
The FT noted that AI companies have prioritized speed to avoid a shortage of computing capacity, even if that meant relying on inexperienced partners.
The FT said some in the industry have likened this to the “Wild West.”
As money pours into startups and due diligence on new projects moves quickly, AI companies are willing to absorb even short-term cost increases.
In a paper for the Brookings Institution, Columbia Business School professor Stijn Van Nieuwerburgh estimated that U.S. AI infrastructure spending from 2025 to 2032 would reach $10.3 trillion, averaging 3.6% of annual GDP. That would be about 1.5 times the peak of the 19th-century railroad boom and more than three times the internet boom.
He warned that the massive debt was “all hidden off the balance sheet” and that equity value could disappear even if cash flow fell by just 10%.

The Information, an IT publication, cited executives in the data center and neocloud industries as saying that the hundreds of neoclouds operating today would be reduced to around 7–10 within a few years.
Two people made that prediction, and several voices across the industry said consolidation was inevitable, the publication reported.
The publication noted that if a data center leasing company such as CoreWeave collapses, the impact on construction itself would be limited because the facilities could easily find new tenants. But if companies that build their own facilities, such as Nebius, IREN and Nscale, falter, it would be a greater burden for Nvidia.
David Driggers, CEO of Cirrascale, an early neocloud, said, “Consolidation will definitely happen,” and predicted that asset managers or telecommunications and fiber-optic cable companies could pursue acquisitions.
[email protected] Lee Seok-woo, International Affairs Specialist Reporter