Stymied datacentre projects threaten global AI revolution
TL;DR
Large datacentre projects are hitting harder limits worldwide: permits, grid connections, construction costs, supply chains and local opposition are slowing or killing plans. Uptime Institute has identified 250 projects announced since 2021 that each need more than 100 MW of power. It expects roughly half to be cancelled or delayed. Virginia’s Prince William Digital Gateway is now uncertain after a court ruling and the withdrawal of a key backer, with opposition tied partly to a nearby Civil War battlefield.
Nauti's Take
Compute is no longer a resource you can assume will be available on demand. Teams putting AI workflows into production should verify regions, quotas, latency, pricing tiers and fallback providers first.
Model quality matters less when the stack becomes fragile under capacity limits, longer waitlists or rising power-linked costs.
Briefingshow
AI growth depends on power, water, land, transformers, chips and permits. If that infrastructure arrives more slowly than model providers promise, compute access, cloud prices and product roadmaps become less predictable. For companies, AI capacity is now an energy and location problem, not just a software issue.
Sources