America's Grid Faces a New, Power-Hungry Tenant: AI

The explosive growth of artificial intelligence is testing the limits of the U.S. power grid. From Virginia to the Midwest, clusters of data centers running complex AI models are demanding electricity at a scale comparable to small cities, pushing aging infrastructure toward a breaking point.

Grid operators are issuing stark warnings. PJM Interconnection, which coordinates power for 65 million Americans, has flagged potential capacity shortfalls, raising the specter of controlled blackouts within this decade to prevent wider collapse. The problem is structural: these facilities require a constant, massive flow of power for computation and cooling, with little ability to reduce consumption during peak hours.

The consequences are already reaching kitchen tables. In areas with dense data center growth, electricity bills are climbing as utilities invest in new infrastructure and compete for fuel. One Bloomberg analysis found wholesale power costs in some regions have soared by over 250% in five years, a surge directly linked to digital expansion.

Projections show no slowdown. Data centers, which consumed about 4% of U.S. electricity in 2024, could more than double their share by 2030. Globally, AI's power appetite might claim 4.4% of the world's electricity by 2035. This demand is colliding with a slow-moving energy sector, where permitting and building new power plants, especially nuclear or renewable projects, takes years.

Some tech companies are responding. Microsoft has offered to pay premium rates in communities where it builds data centers to help offset local price hikes. Yet critics see this as a temporary fix for a systemic issue. The real solution requires modernizing a grid built for a different era, accelerating clean energy deployment that can meet 24/7 demand, and fostering deeper collaboration between utility planners and Silicon Valley. The nation's technological ambition now hinges on keeping the lights on.