Overview
- Late October 2026 analysis and reporting found that grid access and reliable electricity have become the binding constraint on new AI data centres, replacing earlier concerns about chip supply.
- Grid indicators show acute strain with ERCOT’s large-load interconnection queue expanding dramatically to hundreds of gigawatts by mid-2026 and PJM capacity auctions clearing at the regulatory price cap for multiple rounds, signaling shortfalls in available supply.
- Major cloud providers have responded by signing multidecade power deals, including long-term purchases of nuclear output, while the U.S. Department of Energy is financing reactor restarts to speed new supply.
- Those commercial moves are already raising system costs: PJM’s market monitor attributes roughly $29.4 billion in recent capacity-auction costs to projected data-centre demand and advocacy groups warn of sizable household bill increases, a political flashpoint in local races and ballot measures.
- Policy responses remain fragmented as U.S. federal regulators rejected most of PJM’s mandatory-supply proposal and the U.K. offered a non-binding data-centre covenant, leaving decisions on cost allocation, permitting speed and local limits unresolved and tilting firms toward on-site generation, storage and flexibility measures.