“47 U.S. data centers disrupted… AI infrastructure boom cracks”

(Seoul = Yonhap News) Reporter Joo-ho Jeong = The British Financial Times (FT) reported on the 18th (local time) that cracks are appearing in the U.S. artificial intelligence (AI) infrastructure investment boom, estimated to be worth $10 trillion.

While the AI ​​boom has continued for more than three years, data center construction plans are running into realistic problems.

Citing data from regulatory information company Leglayer, the FT reported that at least 47 large data center projects have been delayed, withdrawn, or blocked from approval this year. Legrayer described the figure as a "conservative estimate" that only covers the largest projects.

Earlier this month, investment bank Bernstein also estimated that only about 35% of data center projects planned in the United States were feasible.

The FT added that its research found that more than 1,000 communities across the U.S. had attempted to introduce moratoriums or other restrictions on building AI infrastructure this year.

The financial situation is also worsening.

Neocloud companies, which rent and operate AI data centers, are having difficulty raising funds due to rising interest rates and worsening bond markets.

Coreweave's stock price fell 23.6% last week, and the trading price of one of the corporate bonds issued by Coreweave is in non-performing loan territory.

Softbank affiliate SB Energy and Australia's Permas Grid, in which Nvidia invested, have postponed or put on hold their listing plans.

Several sites being developed by Oracle for 'Stargate', a large data center project of Open AI, Oracle, and Softbank, are also experiencing schedule delays, and there is a growing possibility that Antropic will terminate its $45 billion contract with British neocloud Nscale.

The FT pointed out that AI companies have prioritized speed to avoid a lack of computing power, even if they rely on inexperienced partners.

It is said that there are voices in the industry comparing this to the 'Wild West'.

As funds are pouring into new companies and new project due diligence is progressing rapidly, AI companies are facing short-term cost increases.

In a Brookings Institution paper, Professor Stein Pannierwerch of Columbia Business School estimated that U.S. AI infrastructure spending between 2025 and 2032 will reach $10.3 trillion, accounting for 3.6% of the annual average gross domestic product (GDP). It is about 1.5 times the peak of the railroad boom in the 19th century and more than three times the Internet boom.

He warned that massive debt "is all lurking outside the financial statements" and that even a 10% drop in cash flow could wipe out the value of the shares.

In relation to this, IT media The Information quoted data center and neocloud industry executives and said that within a few years, the number of neoclouds currently numbering in the hundreds will be reduced to about 7 to 10.

Two people made this prediction, and it is said that there were several views across the industry that integration was inevitable.

The media pointed out that even if a company that rents a data center like Coreweave collapses, the damage to the construction itself will be small because it is easy for the facility to find a new tenant, but if companies that build the facility themselves, like Nevius, Irene, and Enscale, are shaken, it will be a greater burden on Nvidia.

David Driggers, CEO of Sirasscale, an early neocloud, said, “Integration will definitely occur,” and predicted that asset management companies or telecommunications and optical cable companies could take over.