US data center power constraints force major project delay or curtailment before mid-2026

AI infrastructure scaling is colliding with physical power delivery constraints. US data center power demand is projected to grow 15-20% annually through 2030, but grid interconnection queues in PJM and ERCOT now stretch 4-5 years, and GSU transformer lead times exceed 24 months. Major operators—Microsoft, Meta, Google, Amazon—have all signaled power procurement as a binding constraint on capacity expansion. Virginia's Loudoun County, the world's largest data center cluster, faces transmission congestion that could force load curtailments. This perpetual market tests whether the AI power bottleneck moves from projection to confirmed disruption. LONG positions benefit from each confirmed project delay or curtailment attributed to power constraints; SHORT positions benefit if utilities and operators successfully accelerate grid upgrades and keep pace with demand growth. Key evidence sources: FERC interconnection queue reports, utility integrated resource plans, data center company 10-K/10-Q risk factor disclosures, transformer manufacturer order books, and state utility commission dockets on data center load growth.