Overview
- Researchers and grid operators say electricity and grid access have overtaken semiconductors as the main constraint on AI buildout because new power plants and transmission take years to deliver.
- Texas illustrates the pressure on grids: ERCOT’s large-load queue jumped from about 63 gigawatts to roughly 474 gigawatts with most requests tied to data centers, and PJM capacity auctions recently cleared at price caps while failing to meet reliability targets.
- NERC forecasts roughly 224 gigawatts of higher summer peak demand over the next decade and warns many regions could face shortfalls by 2030, signaling a sustained mismatch between where AI needs power and where it is available.
- Markets and policy proposals now center on faster permitting and queue reform, plus near-term flexibility tools such as battery storage, demand response and on-site generation to free capacity while major grid upgrades proceed.
- Analysts expect a large investment wave to follow the power race, with forecasts of sharp rises in data-center electricity demand and trillions in capital spending that will drive who pays for new infrastructure and where jobs and investment land.