At least 47 major US data centre projects halted or cancelled

At least 47 large data centre projects have been halted or cancelled across the US, as a massive $10 trillion investment wave in artificial intelligence infrastructure begins to stumble amid construction delays, rising borrowing costs, and local public pushback.

According to a report by the Financial Times, this emerging picture threatens the future of hundreds of projects while deepening investor doubts over returns on the colossal sums being spent.

Data from research firm RegLayer shows that at least 47 major data centre projects across the US have been delayed, cancelled, or brought to a complete standstill.

In an assessment published in October, analysts at financial firm Bernstein pointed out that only about 35% of the projects planned across the country can be considered realistic.

Local authorities and the bond market apply the brakes

The obstacles facing investments are not limited to technical plans. More than 1,000 localities across the US have sought to impose moratoriums or other restrictions on AI infrastructure construction this year.

The sector’s growth depends directly on whether next-generation cloud providers, often referred to as “neoclouds”, can secure financing for these giant buildouts.

However, high interest rates, inflation, and disruptions across the bond market have shaken investor confidence in the ability of companies such as OpenAI and Anthropic to generate revenues that cover their expenses.

This loss of market confidence prompted investors to rapidly exit debt securities tied to AI infrastructure companies. Shares in cloud provider CoreWeave plunged by around 25% in a single week.

Similarly, SoftBank unit SB Energy and Nvidia-backed Firmus delayed or suspended their plans to go public. OpenAI also indicated that it deferred its own initial public offering, citing safety and security reasons.

Delays hit giant Stargate project

Difficulties on the ground are directly affecting construction sites. Delays have hit several Oracle sites tied to the $500 billion Stargate project, which OpenAI is pursuing to build data centres.

These schedule slippages could grant customers the right to walk away from contracts if facilities are not completed on time, while contractors could demand additional financing.

Investments could exceed $10 trillion

Despite the mounting challenges, the volume of capital flowing into the sector remains at record levels.

Columbia Business School economist Stijn Van Nieuwerburgh calculates that US AI infrastructure spending between 2025 and 2032 could exceed $10 trillion, equivalent to roughly 3.6% of the country’s gross domestic product each year.

As investors weigh risks far more strictly, obtaining building permits for data centres and connecting facilities to the power grid have grown increasingly difficult.

In a report published in late July, the Financial Times noted that Google, Amazon, Microsoft, and Meta had collectively invested $1.1 trillion since the AI boom began in 2023.

Big tech companies plan to allocate $745 billion this year toward data centre construction, advanced chip purchases, and building computing capacity.

Returns on these investments hinge on whether OpenAI and Anthropic can generate the resources needed to cover their computing costs.

Although spending has boosted corporate revenues, particularly in the cloud sector, supply chain strain has triggered memory chip shortages, adversely affecting even players that have fallen behind in the AI race, such as Apple.