Starting June 2027, PJM Interconnection — the largest US grid, serving 67 million customers from Virginia to Illinois — will cut power to data centers drawing more than 50 megawatts (a class that covers most new hyperscale builds) before dimming
PJM Interconnection, the largest US grid operator, will cut power to large data centers before dimming household lights when demand outruns supply on its 13-state network, the grid operator announced this week. The mechanism takes effect in June 2027 and applies only to data centers drawing 50 megawatts or more, a class that covers most new hyperscale builds.
This is the first time a major US grid has formally named data centers as a first-cut curtailment class, a category previously reserved for aluminum smelters and other large industrial customers. Curtailed operators will be paid for the lost load and warned between 30 minutes and a few days in advance, mirroring long-standing demand-response programs in which big users volunteer to power down in exchange for compensation when the grid is strained.
PJM's territory runs from Virginia to Illinois and serves 67 million customers. Data centers are expected to use roughly four times as much electricity by 2035 as they do today, according to industry projections cited by TechCrunch. PJM's 2026 capacity auction, the annual market in which generators are paid to be available a year ahead, fell short of the new generation the grid needed, framing the curtailment move as a stopgap rather than a permanent answer.
Wholesale electricity prices in PJM territory have nearly doubled over the past year, and the grid's independent market monitor has blamed data center load for much of the increase, per TechCrunch's reporting. A new policy that asks hyperscale operators to act as an industrial demand-response class is the grid operator's way of buying time until the next auction, or the next interconnect, can close the gap.
The fallback is already in motion, and it has an air-quality cost. Federal rules let diesel backup generators run up to 50 hours a year for demand-response events and up to 100 hours a year for emergencies and maintenance. When a data center loses its grid feed, the on-site generators carry the load. A report on a 96-megawatt Vantage Data Centers site in Northern Virginia estimated tens of millions of dollars in annual health damages from diesel exhaust, and Vantage has drawn criticism for allegedly coordinating with Virginia environmental regulators to discredit the report. The DOE's Emergency Order 202-26-23 and the PJM 202(c) backup-generation application give the operator an emergency lever to curtail large loads during hot-weather events, and that lever points the same way: more curtailments, more diesel runtime, more local pollution where the data centers sit.
The expected response is on-site power: gas turbines, fuel cells, behind-the-meter solar, or long-duration batteries paired with a small nuclear or gas plant. Operators with their own generation can offer the grid reliability they used to consume, and that flips the politics of the next capacity auction.
The honest falsifier is the capacity auction. If the 2026 auction had cleared enough new generation, this policy would not exist. The June 2027 mechanism is what an operator does when the buildout is faster than the wires, transformers, and turbines that have to follow it. PJM's bet is that paying data centers to step aside is cheaper than paying for a fleet of plants the grid will not finish in time.