The Numbers Are Hard to Ignore
Power prices on PJM Interconnection — the massive grid serving over 65 million people across 13 U.S. states and Washington, D.C. — have spiked 76% in recent months. The culprit, according to a federal energy watchdog, isn't a shortage of fuel or a natural disaster. It's data centres.
The explosive growth of artificial intelligence has sent electricity demand soaring across North America, and the U.S. grid is struggling to keep up. The watchdog's report is a stark reminder that the infrastructure underpinning the modern economy was designed for a very different world.
A Grid Built for the Past
The American electricity grid, in most regions, was engineered decades ago to handle residential, commercial, and traditional industrial loads. It was not built to absorb the power appetite of thousands of GPU clusters running large language models 24 hours a day.
AI training and inference workloads are uniquely demanding. Unlike a factory that runs one shift or a shopping mall that dims its lights at midnight, hyperscale data centres draw enormous, near-constant loads. When dozens of new facilities come online in the same regional grid within a short window — as has happened across Virginia, Ohio, and other PJM states — the math stops working.
The result: generators and utilities are charging a premium for power that's increasingly scarce at peak times, and those costs are beginning to ripple outward.
Who Gets Blamed?
The watchdog report doesn't let any single party off the hook. Grid planners are criticized for underestimating how fast AI infrastructure would scale. Regulators are faulted for slow-walking transmission approvals. And tech companies themselves face scrutiny for signing power agreements faster than the grid can accommodate them.
At the heart of the issue is a mismatch in timelines: a new data centre can be planned, financed, and partially built in two to three years. A new high-voltage transmission line — or a grid-scale energy storage project to buffer demand — can take a decade or more to permit, build, and connect.
The Wider Stakes
The price spike on PJM isn't just a regional American problem. Canada's grid operators are watching closely, since continental electricity markets are increasingly interconnected. Higher U.S. power prices can pull Canadian exports southward, tightening domestic supply. Ontario and Quebec, both significant electricity exporters, are already fielding more aggressive inquiries from U.S. buyers hungry for clean, reliable power.
There's also the climate dimension. When grid capacity tightens, utilities often lean on the fastest-available generation — which tends to be older, dirtier peaker plants. The AI boom's carbon footprint isn't just about the chips; it's about what burns to power them when the sun isn't shining and the wind isn't blowing.
What Comes Next
Industry analysts expect the demand crunch to intensify through the late 2020s as AI adoption accelerates across sectors. Solutions on the table include accelerated permitting for new transmission, incentives for on-site renewable generation at data centres, and demand-response programs that could briefly curtail non-critical AI workloads during peak strain.
For now, the 76% price increase is a signal the grid itself is trying to send: the infrastructure of the 20th century is being asked to carry the ambitions of the 21st, and the gap is widening fast.
Source: TechCrunch
