Business Korea · সাধারণ
U.S. Data Center Boom Shows Cracks
১৯/১০/২০২৬

Data center.
In the United States, the world’s largest artificial intelligence (AI) infrastructure market, at least 47 large-scale data center projects have faced disruptions this year alone. Concerns are rising that cracks are forming in the U.S. AI infrastructure boom, estimated to be worth $10 trillion, due to a combination of practical issues such as worsening corporate financial conditions, power shortages, and opposition from local residents.
On Oct. 18 (local time), the Financial Times (FT) cited data from regulatory intelligence firm RegLayer, reporting that at least 47 data center development projects in the U.S. have been delayed, withdrawn, or halted this year. RegLayer explained, “We surveyed primarily large-scale worksites, and the figures were compiled conservatively.” Recently, investment bank (IB) Bernstein also estimated that only 35% of the data center construction projects currently planned in the U.S. are feasible.
Funding shortages are cited as the primary cause of the successive disruptions in AI infrastructure projects in the U.S. Concerns have grown over the structure in which new operators take out loans to bear massive costs for purchasing GPUs and securing sites. In particular, the controversy intensified recently as the corporate bond yield of AI infrastructure company CoreWeave surged to the 15% level. CoreWeave is a so-called ‘neocloud’ company that operates a business of buying Nvidia AI chips to build data centers and then leasing the infrastructure to tech companies. The FT stated, “The overheating of infrastructure competition among companies has turned the overall AI industry into a ‘Wild West’,” adding, “Interest rate hikes and the bond market slump are threatening the ambitions of AI infrastructure companies.”
The increasingly severe power shortage and opposition from local residents, who have begun to perceive data centers as undesirable facilities, are also stumbling blocks. Big tech company Oracle notified the developer of the 2.45GW (gigawatt) ‘Project Jupiter’ underway in New Mexico of force majeure last month in preparation for disruptions in securing power. The intent is to delay rent payments if the set schedule is not met. PJM, the largest power grid operator in the US, recently decided to halt a previously planned special auction for 15 years’ worth of power for data centers to respond to the AI-driven power shortage.
According to the FT’s own survey results, it was found that more than 1,000 communities in the US have attempted to introduce temporary halts or restrictive measures related to AI infrastructure development this year.
Concerns over AI infrastructure companies are emerging in various parts of the market beyond the US. It is reported that Nscale, a British neocloud company, is also pursuing an initial public offering (IPO) with a corporate value of about $30 billion, but is facing difficulties as investors’ expectations are much lower than this. Bloomberg reported that due to the recently deteriorated industry conditions, some investors view Nscale’s corporate value at $16 billion to $19 billion. Previously, Australian AI infrastructure company Firmus also withdrew its IPO plan, which had been pursued at a scale of about $5 billion.
It is pointed out that such crises for AI infrastructure companies could ultimately affect the earnings of semiconductor manufacturers. This is because if infrastructure companies increasingly struggle to secure investment funds and construction plans are consequently delayed, they may delay or reduce orders for semiconductor chips to be used in data centers.
In fact, the stock prices of domestic memory companies such as Samsung Electronics and SK hynix have continued to show weakness recently. This is due to worsened investor sentiment caused by concerns over U.S. interest rate hikes and the pace adjustment of AI infrastructure. Bloomberg stated, “As borrowing costs rise and uncertainty about the long-term profitability of AI technology grows, investor sentiment regarding overall AI infrastructure, including server companies and semiconductor manufacturers, has deteriorated.”
মূল প্রতিবেদনে যান ↗