The PJM Interconnection, the grid operator that coordinates electricity for 65 million people across thirteen states and the District of Columbia, issued an unusually blunt warning this week: the region has "years, not decades" to overhaul the way it manages power. The catalyst is not a sudden coal-plant retirement or a heat wave. It is the relentless growth of cloud computing and artificial intelligence.
In a 70-page white paper, PJM CEO David Mills wrote that "the current situation is not tenable." The operator's territory includes Northern Virginia, the world's densest concentration of data centers, where new AI training clusters and hyperscale cloud campuses are coming online faster than the grid can absorb them. The strain is so acute that American Electric Power, one of the nation's largest utilities, is publicly considering a pullout from PJM altogether. "If something is not done now, I expect we could still be having these same conversations in ten years," AEP CEO Bill Fehrman told investors.
The roots of the crisis stretch back to 2022, when PJM paused new interconnection requests to clear a years-long backlog. At the time, more than 300 gigawatts of generation projects were waiting to connect—roughly triple the grid's current peak load. Only 23 gigawatts have actually made it online since then, and most developers withdrew rather than endure the wait. Since reopening the queue, PJM has already received more than 800 new requests totaling 220 gigawatts, suggesting demand is accelerating, not cooling.
PJM's proposed fixes boil down to three options: force utilities and generators into longer-term supply commitments; create tiered reliability guarantees that let lower-paying customers get cut first; or shift closer to a real-time pricing market. None of the scenarios are politically easy, and all could raise costs for consumers.
Why it matters
Data centers are the new factories, and electricity is the new raw material. If the largest U.S. grid cannot align its planning process with the speed of AI infrastructure deployment, the bottleneck will move from silicon to electrons. For enterprise CIOs and cloud buyers, this is a supply-chain risk that deserves the same scrutiny as cybersecurity or vendor lock-in.