America’s $10 trillion artificial intelligence (AI) infrastructure boom is facing growing problems. Data centre projects are being delayed, some are being cancelled or blocked, and investors are becoming less confident about the sector. These challenges are raising questions about whether the massive spending planned for AI infrastructure can continue at the expected pace.
The AI boom has created a market that some industry participants describe as the “Wild West”. Over the past few years, companies rushed to build data centres to meet the growing demand for computing power. According to people involved in the market, huge amounts of money flowed into newly established companies, while checks on new projects were sometimes completed more quickly than usual, as reported by Financial Times.
US AI data centre projects face delays
Many data centre projects are now struggling to move forward. More than three years after the AI boom began, ambitious construction plans are facing practical challenges. At least 47 large data centre projects were delayed, withdrawn or blocked in 2026, according to data shared by RegLayer, a regulatory intelligence company. RegLayer described this figure as a conservative estimate because it focused on the largest projects.
Only a fraction of planned US data centre projects may be realistic. Analysts at Bernstein estimated this month that only about 35% of the planned US data centre project pipeline was credible. This raises doubts about how many of the announced projects will actually be completed.
{{/usCountry}}Only a fraction of planned US data centre projects may be realistic. Analysts at Bernstein estimated this month that only about 35% of the planned US data centre project pipeline was credible. This raises doubts about how many of the announced projects will actually be completed.
{{/usCountry}}Local opposition is creating another hurdle for data centre developers. More than 1,000 communities across the US have tried to introduce temporary bans or other restrictions on AI infrastructure in 2026, according to Financial Times research. This opposition is making it harder for companies to obtain the approvals needed to build new facilities.
Higher interest rates and a weaker bond market are threatening the expansion plans. Several people involved in the AI infrastructure market said rising borrowing costs and deteriorating conditions in the debt market could make it harder for neocloud companies to raise the money needed to build data centres. This could slow down the wider AI infrastructure boom.
Also read: Wall Street AI rally faces pressure ahead of Meta, Tesla and major chipmakers' earnings reports
CoreWeave stock falls as investors worry
Investors have started reassessing the risks of AI infrastructure companies. During the week, debt linked to several AI infrastructure businesses, including cloud computing provider CoreWeave, fell sharply. Shares of listed companies in the sector also declined. CoreWeave's stock lost almost 25% of its value during the week, according to Financial Times.
Several companies have delayed or put on hold their plans to go public. New data centre developers and cloud providers, including SoftBank's SB Energy and Nvidia-backed Firmus, have delayed or shelved their planned initial public offerings (IPOs). OpenAI's decision to delay its own IPO has also weakened investor confidence in the sector.
Construction delays could allow customers to walk away from data centre agreements. Large technology companies and AI labs typically sign contracts to lease computing capacity from data centre operators. Some agreements contain clauses that allow customers to cancel their commitments if facilities are not delivered within the agreed timelines. Delays could therefore leave developers with expensive projects and uncertain future income.
Data centre construction costs rise
Former cryptocurrency mining companies are also adjusting their AI infrastructure plans. TeraWulf and Cipher, which previously operated in cryptocurrency mining, have renegotiated leases and extended delivery schedules slightly for projects intended to serve Anthropic. The changes were partly linked to design adjustments made by Anthropic and its partner Fluidstack in recent months, according to Financial Times.
Cipher has agreed to absorb a large part of the additional construction costs at a Texas project. When Cipher amended its lease agreement with Fluidstack last month, it secured more time to deliver the facility but agreed to cover the first $359 million in cost overruns. The changes required approximately $750 million in additional equity funding, partly supported by an $850 million convertible bond raised this week.
Anthropic could cancel its $45 billion agreement with Nscale if the company misses key targets. Nscale, a London-headquartered neocloud launched in 2024, has a major project planned in West Virginia. Anthropic can walk away from the agreement if Nscale fails to meet its financing or delivery deadlines.
Also read: Microsoft pulls out of Nscale data center deal in UK amid power supply delays, report says
Nscale's plans to go public could face difficulties because of unstable markets. The Nvidia-backed company is targeting a New York stock market listing at a valuation of $35 billion. However, bankers said volatile market conditions would make the proposed IPO difficult to complete in the near term. A person familiar with the matter said Nscale hoped to list this year. Nscale declined to comment.
Nscale still needs to expand its computing capacity significantly to meet its ambitions. The company had only 25,000 advanced AI chips operating as of August, compared with nearly 500,000 chips it plans to have in operation. This gap highlights the scale of the expansion it must complete, according to Financial Times.
Nvidia chief executive Jensen Huang expressed confidence in Nscale's growth. Speaking in London on Wednesday, Huang said Nvidia had helped build up the company within a year. He described Nscale as a promising British start-up and suggested it could become the most valuable UK company created in the previous 10 years.
US AI infrastructure spending may top $10 trillion
The planned spending on US AI infrastructure is unprecedented in scale. US spending on AI infrastructure is projected to exceed $10 trillion between 2025 and 2032, according to estimates by Stijn van Nieuwerburgh, an economist at Columbia Business School, in a paper for the Brookings Institution.
CoreWeave has become a key example of the financial risks facing the sector. The company was a relatively small cryptocurrency miner four years ago before taking advantage of rising AI demand. It has since become a major provider of AI computing infrastructure, but investors are increasingly concerned about its ability to finance billions of dollars in Nvidia hardware purchases and data centre leases, according to Financial Times.
CoreWeave's debt has come under pressure as investors reassess its financial position. One of its unsecured bonds fell into distressed territory during the week. Bonds financing data centres leased to CoreWeave also declined in value. A $1.1 billion bond linked to a data centre site in Virginia lost about 5% of its value in the three weeks after it was priced.
SB Energy faces enormous funding requirements despite having no completed data centre. The company is majority-owned by SoftBank, led by Masayoshi Son, and is backed by OpenAI and Nvidia. It has not completed a single data centre but needs $174 billion in capital to meet its commitments under contracts that could generate $430 billion in future revenue, largely from OpenAI.
SB Energy has delayed its planned stock market listing. The company had been considering an IPO at a valuation of $50 billion. However, its listing plans have been postponed as financial and market conditions become more challenging, according to Financial Times.
Nebius has also highlighted the enormous cost of expanding AI infrastructure. Arkady Volozh, chief executive of Nvidia-backed neocloud company Nebius, told a Citi conference in September that building its planned 5-gigawatt capacity would require about $250 billion in financing. He said the company had raised financing for only around 20% of that amount so far, according to Financial Times.
AI data centres face power and permit hurdles
Access to electricity is another major challenge for developers. Utility regulators have been refusing grid connections to some proposed data centres unless the projects can provide their own power, according to data centre executives. This requirement could force companies to find additional energy sources and spend more money before construction plans can proceed.
OpenAI's $500 billion Stargate infrastructure plan has faced several setbacks. The project was intended to support Altman's AI ambitions while aligning them with US President Donald Trump's technology agenda. However, it has encountered problems including rejected or blocked planning permissions, construction delays and the departure of senior executives.
The financial environment for data centre developers has become much tougher. An industry executive said investors and lenders were no longer willing to provide money for data centre projects as easily as before. The executive said financing such facilities had become significantly more difficult, highlighting the change in sentiment across the sector, according to Financial Times.
The central challenge for America's AI boom is turning huge investment plans into working infrastructure. Although AI companies expect computing demand to keep growing, developers must still secure financing, complete construction, obtain regulatory approvals and provide reliable power. Delays, rising costs and weaker investor confidence could threaten the pace of expansion and make it harder for companies to deliver on their ambitious AI plans.