Overview
- By late October 2026 analysts and grid authorities concluded that access to timely, local electricity — not semiconductors — now determines where and how fast new AI data centers can come online.
- Grid metrics show strain: ERCOT’s large‑load queue ballooned from about 63 GW at end‑2024 to roughly 474 GW by mid‑2026, PJM capacity auctions cleared at the price cap and missed reliability targets, and NERC forecasts a roughly 224 GW rise in summer peak demand that puts many regions at shortfall risk by 2030.
- Political pressure is rising as protests, ballot measures and tighter state rules have stalled or delayed projects in places such as Pennsylvania and parts of the UK, while the UK’s new data‑centre covenant and reported withdrawals from planned sites reflect growing local pushback and power delays.
- Industry and policy responses now focus on near‑term fixes that can free capacity faster than yearslong grid builds — battery storage, demand response, on‑site generation, staged connections and queue reform — even as analysts warn trillions of dollars in new capital will be needed to fully close the gap.
- The rush for power carries mixed local effects: some towns see big tax revenue and public spending gains from data centers, others face higher housing and infrastructure costs, and economists warn the buildout may crowd out construction in housing and manufacturing if capital and labor remain tied up.