The AI build-out has slammed into the physical world. New analysis suggests that 30 to 50 percent of roughly 140 planned U.S. data centers — together targeting about 16 gigawatts of capacity — may miss their 2026 timelines or be canceled outright.
The bottlenecks are mundane and stubborn
The culprits are not exotic. They are transformers with multi-year lead times, batteries, grid interconnection queues, and local opposition to enormous power-hungry facilities. Each of these is hard to fix quickly, and together they form a wall that no amount of AI investment can simply wish away.
Why the money can't just solve it
There is no shortage of capital chasing AI infrastructure — if anything, the headlines are full of multi-billion-dollar raises. But money cannot manufacture a transformer faster than the supply chain allows, nor compress the years it takes to connect new load to a strained grid. Power and hardware, not funding, are the binding constraints of this cycle.
Why it matters
This is the unglamorous flip side of the AI boom. Every projection of explosive compute growth implicitly assumes the data centers actually get built on schedule — and a large share of them may not. For anyone planning around abundant, cheap compute, the prudent assumption for the next couple of years is that capacity will arrive later, and cost more, than the roadmaps suggest.
