The electricity bill for the AI buildout has arrived. It just came addressed to the grid first. New market data from PJM Interconnection, which manages wholesale power across 13 US states and the District of Columbia, shows total wholesale electricity costs averaged $136.53 per megawatt-hour in the first quarter of 2026, up from $77.78 a year earlier. That is a 75.5 percent increase in a single year, and it happened in the region directly beneath the largest concentration of AI data centers in the world.
PJM sits under Northern Virginia. The county of Loudoun alone hosts more data center capacity than most countries. When hyperscalers, cloud providers, and AI infrastructure companies build out compute at the scale the last two years have demanded, the power grid is not a passive backdrop. It is a finite physical resource with hard constraints on how fast new capacity can be added.
The most revealing number in PJM's latest market monitor report is not the energy price itself. It is the capacity price, which rose 398 percent year-over-year, from $3.57 to $17.78 per megawatt-hour. Capacity markets do not price what electricity costs today. They price what the grid needs to pay in order to guarantee that enough generation will be available when demand peaks in the future. A near-quadrupling of capacity costs means the market is already repricing anticipated future demand, including data center loads that have not yet come fully online.
This matters because it locks in higher costs before facilities are even built. Once transmission projects are planned, capacity obligations are set, and new load forecasts enter the market, the cost structure changes for years. Monitoring Analytics, PJM's independent market monitor, has already described the impact of data center growth on regional prices as significant and irreversible. That is not typical language from a market regulator.
Most AI companies do not buy wholesale electricity from PJM directly. They access compute through AWS, Microsoft Azure, Google Cloud, or colocation providers, which are the entities actually negotiating power contracts and managing grid connections. But their costs eventually flow through pricing. Cloud rates are sticky and competitive, protected by long-term contracts, and the market is genuinely global, which provides some insulation. Still, direction is hard to ignore when a dominant grid region sees 75 percent price increases in twelve months.
The deeper structural issue is that AI infrastructure is concentrating in exactly the places where grid constraints are most severe. Northern Virginia's power problems are well-documented: interconnection queues stretching for years, limited substation capacity, and transmission corridors that were not designed for the density of load being proposed. AI companies have been circling other regions for this reason, including the Midwest, the Southeast, and internationally. But the economics of latency and existing fiber density keep pulling development back toward the established hubs.
The AI conversation of the last three years has been mostly about models, chips, and software. Those are the layers where the most visible competition plays out. But the electricity grid is becoming a constraint that no amount of model efficiency improvement can dissolve. The cost of a token is not just the cost of the GPU that computed it. It includes the cost of keeping that GPU powered, cooled, and connected to a grid that is now pricing its own scarcity.