AI Data Centers Drive 76% Electricity Surge in PJM Grid Through 2028

Wholesale electricity prices in the PJM Interconnection—the largest regional power grid in the United States, spanning Virginia, Ohio, Pennsylvania, New Jersey, Maryland, and Delaware—jumped 76 percent in the first quarter of 2026. The culprit is not a temporary spike but a structural shift: artificial intelligence data centers operated by hyperscalers like Microsoft, Amazon, Google, and Meta are consuming electricity faster than utilities can generate it. According to Monitoring Analytics, the independent market monitor for PJM, this pressure is locked in through 2028 via capacity contracts already signed. For founders building AI infrastructure, this means something concrete: the cost of training and running inference on models could nearly double within two years, especially for those relying on Northern Virginia's Data Center Alley, the world's densest concentration of computing facilities.

The problem is structural, not temporary. Demand from AI data centers is growing faster than new generation capacity can come online. When hyperscalers bid aggressively in capacity auctions, they drive up prices for everyone on the grid—utilities, manufacturers, households, and smaller tech companies alike. Monitoring Analytics flagged three critical issues in its 2026 report: additional data center demand outpaces new firm energy supply; fossil fuel plant closures combined with transmission bottlenecks and interconnection delays are raising the risk of sustained high capacity prices; and real concern exists about whether the system can handle new loads before 2028. The impact is not cyclical. Even if AI demand stabilized tomorrow, prices would remain elevated because of contracts already locked in.

The scale of electricity consumption tells the story. The International Energy Agency reported that data center electricity use grew 17 percent in 2025, but AI-focused facilities advanced nearly 50 percent in the same period. By 2030, total data center consumption could nearly double from 485 terawatt-hours to 950 terawatt-hours. AI-dedicated centers alone could triple their consumption. Globally, data centers consumed roughly 415 terawatt-hours in 2024, representing about 1.5 percent of worldwide electricity. In the United States specifically, data centers consumed 176 terawatt-hours in 2023—4.4 percent of the national total—and could reach 325 to 580 terawatt-hours by 2028, potentially accounting for up to 12 percent of all U.S. electricity. Goldman Sachs Research projects data center energy demand will grow 165 percent before 2030, driven almost entirely by generative AI.

Social resistance is mounting. Communities in Northern Virginia, Texas, Arizona, Georgia, Ohio, and Pennsylvania are pushing back against new data center proposals. Local concerns center on rising electricity bills, intensive water consumption for cooling systems, land use with minimal job creation, and fears that grid strain will drive up residential rates. Several jurisdictions have imposed temporary moratoriums on new approvals, pausing development to update energy planning. For AI startups, this means fewer locations to deploy infrastructure and greater concentration of computing power among well-capitalized players who can negotiate long-term power purchase agreements.

For founders, the implications are direct. If your model runs 24/7 inference in congested regions like PJM, expect indirect energy cost increases of 50 to 100 percent through 2027 and 2028 via cloud providers passing through their own rising costs. Regulatory moratoriums shrink the menu of deployment locations. But well-capitalized startups that lock in multi-year power contracts now will have a competitive edge over those dependent on spot market pricing. The concrete moves: diversify deployment across regions with energy surplus rather than concentrating in Northern Virginia; negotiate multi-year cloud contracts before 2027 price adjustments hit; implement intelligent scheduling to shift training to off-peak hours, potentially cutting energy costs 30 to 40 percent; optimize models for efficiency through quantization and pruning; and explicitly evaluate colocation providers like Equinix and Digital Realty that have signed renewable power agreements.

Smaller startups face a disadvantage because they lack the volume to negotiate favorable power purchase agreements the way hyperscalers do. The winning strategy combines multi-region cloud deployment, purchasing compute during low-cost periods, and designing more efficient models. Some founders are spotting opportunity in the crisis itself: new categories of startups are emerging around AI workload energy optimization, real-time compute orchestration based on electricity prices, data center energy procurement platforms, efficient cooling solutions, and GPU capacity marketplaces. For companies operating in these spaces, timing is optimal. Demand for energy-aware AI infrastructure is growing exponentially, and the window to capture market share is now.