Nearly half of all US data centers planned for 2026 have been delayed or canceled.
Out of 12 GW of AI data center capacity announced for this year, only about 5 GW is actually under construction. The rest is stalled. The bottleneck isn't money. Alphabet, Amazon, Meta, and Microsoft have committed over $650 billion in AI infrastructure spending. The bottleneck is power.
The US power grid can't keep up. Lead times for high-voltage transformers have stretched from 12-18 months pre-2020 to 3-5 years today. A single missing transformer or piece of switchgear can halt an entire project. Data centers are being built in 18 months but the electrical infrastructure they need takes 60 months to deliver.
Tariffs are making it worse. A significant share of power delivery equipment comes from China. With tariffs adding 15-25% cost increases on critical components, projects that were already tight on budget are getting pushed back or killed entirely.
The pipeline gap extends far beyond this year. For 2027, only 6.3 GW of 21.5 GW announced has broken ground. Through 2032, more than 50 GW of announced capacity lacks firm completion dates. Even OpenAI's $500 billion Stargate project has seen no significant physical progress as of April.
Community opposition is growing too. Residents in Virginia, Georgia, and Texas are pushing back on water usage, noise, and local grid strain. Projects are being canceled at the county level.
The AI demand is real. The infrastructure to support it is not materializing fast enough. Capital cannot outrun physics.
Out of 12 GW of AI data center capacity announced for this year, only about 5 GW is actually under construction. The rest is stalled. The bottleneck isn't money. Alphabet, Amazon, Meta, and Microsoft have committed over $650 billion in AI infrastructure spending. The bottleneck is power.
The US power grid can't keep up. Lead times for high-voltage transformers have stretched from 12-18 months pre-2020 to 3-5 years today. A single missing transformer or piece of switchgear can halt an entire project. Data centers are being built in 18 months but the electrical infrastructure they need takes 60 months to deliver.
Tariffs are making it worse. A significant share of power delivery equipment comes from China. With tariffs adding 15-25% cost increases on critical components, projects that were already tight on budget are getting pushed back or killed entirely.
The pipeline gap extends far beyond this year. For 2027, only 6.3 GW of 21.5 GW announced has broken ground. Through 2032, more than 50 GW of announced capacity lacks firm completion dates. Even OpenAI's $500 billion Stargate project has seen no significant physical progress as of April.
Community opposition is growing too. Residents in Virginia, Georgia, and Texas are pushing back on water usage, noise, and local grid strain. Projects are being canceled at the county level.
The AI demand is real. The infrastructure to support it is not materializing fast enough. Capital cannot outrun physics.
5
