Overview
- Delphi Digital’s October 29 report reframes the bottleneck for AI expansion as electricity and grid access rather than chip supply, arguing that generation, transmission and interconnection take years to deliver while compute can scale faster.
- Grid data show the strain: ERCOT’s large-load queue ballooned from 63 GW at end-2024 to 474 GW by June 2026 with most driven by data centers, PJM’s recent capacity auctions cleared at the price cap and NERC forecasts a roughly 224 GW rise in peak summer demand that leaves many regions at risk by 2030.
- Connection queues are a global problem with an estimated 2,500 GW of projects waiting for grid hookups and European data-center hubs sometimes facing multi-year waits that can exceed a facility’s build time.
- Industry and policy proposals now focus on faster permitting and queue reform, plus short-term fixes such as battery storage, demand response, on-site generation and data-center energy management to free capacity while long-lead transmission and plants are built.
- The power constraint has big economic and political stakes because Goldman Sachs sees data-center demand rising sharply and trillions in capex ahead, raising questions about who finances grid upgrades, how costs are allocated, and the local impacts on rates, jobs and national competitiveness.