Two-thirds of US data center power demand unlikely to materialize

⚡ AI Executive Summary

Wood Mackenzie analysis indicates that utilities and grid operators will likely commit to only about 28% of the 1,066 gigawatts requested for artificial intelligence data center projects across the United States. The disparity highlights a significant gap between stated demand and realistic buildout capacity, driven by speculative applications and duplicate project filings designed to secure favorable terms from multiple utilities simultaneously. This mismatch creates substantial planning challenges for grid operators and threatens to drive infrastructure upgrade costs onto ratepayers. The bottleneck reveals deeper tensions in the energy transition: while rapid AI buildout requires aggressive grid expansion, inflated demand forecasts and inefficient permitting processes risk wasting capital on infrastructure that may never be fully utilized. Regulators and utilities are beginning to implement screening mechanisms and upfront cost requirements to distinguish genuine projects from speculative filings, yet the risk remains that overly stringent barriers could slow legitimate high-value development and undermine US competitiveness in global AI markets.