The Stargate campus in Abilene, Texas — a $40 billion OpenAI and SoftBank joint venture — will draw 1.2 gigawatts at peak load, equivalent to 313,000 U.S. homes. Epoch AI projects global AI compute hitting 100 gigawatts by 2030 if current growth holds. Yet the primary constraint isn't generation capacity but the interconnection queue: median wait times for new plants jumped from under 20 months in 2005 to 55 months by 2023. Grid operators study each request sequentially under rigid first-come-first-served rules that don't prioritize high-value projects or reward flexible load behavior.

Congestion costs in U.S. wholesale markets reached $11.5 billion in 2023, up 45 percent year over year. ERCOT, serving 90 percent of Texas load including Abilene, forecasts insufficient capacity by summer 2028. PJM, the nation's largest grid, failed to procure enough capacity for 2025 demand. MISO warns of growing resource adequacy risks. Meanwhile, electrification economics compound pressure: electric motor prices fell 97.5 percent since 1990, battery costs dropped 98.8 percent, and EVs now cost half as much per mile to operate despite higher electricity rates.

Market signals are already reshaping the generation mix. In ERCOT, solar output surged from 4 to 29.8 gigawatts between December 2020 and September 2025, pushing midday power values from 93 percent to 39 percent of peak-hour prices. That price collapse triggered a storage response: battery discharge capacity grew from near zero to 8.6 gigawatts by October 2025. The mechanism works — developers build what prices demand — but only if interconnection reform lets them connect.