PJM Grid Operator Puts AI Data Centers Last in Line for Power

Grid operator responds to federal mandate with tough new requirements

PJM Interconnection, the grid operator serving 67 million people across 13 states, has proposed rules that would make AI data centers the first to lose power during grid shortages unless they can prove they've secured new, dedicated electricity supply by March 2027.

The proposal responds to a June 18 federal order requiring all six U.S. regional grid operators to justify or rewrite their rules for connecting very large customers. PJM filed its response on August 13, according to Forbes, which first reported the details. The same mandate applies to operators covering the Midwest, Plains, California, New England, and New York, making PJM's approach a likely template for national policy.

The 50-megawatt threshold and interim service

Under PJM's proposal, any new facility drawing 50 megawatts or more at a single site—roughly a mid-sized data center campus—must demonstrate it has secured new power generation capacity. Without that proof, facilities receive what PJM calls "interim service."

Interim service doesn't mean automatic disconnection. During grid shortages, PJM would reduce power only by the amount needed, only in the affected zone, and only for the duration of the shortage. The financial penalty is more telling: traditional customers who reduce load during emergencies receive full compensation rates, while interim service customers receive half that amount. PJM's proposal suggests these customers might waive even that reduced payment.

The policy stems from concrete grid capacity problems. PJM's most recent capacity auction hit its price ceiling and still fell 6,831 megawatts short—the second consecutive auction to miss its target.

New power must actually be new

PJM will accept several paths to qualification: building a new gas plant, upgrading an existing facility, restarting a retired plant, or agreeing to demand reduction when requested. The critical requirement is that the electricity must be new to the grid, not simply purchased from existing sources.

The timeline creates severe practical constraints. Facilities must submit evidence by March 1, 2027, complete final verification by April 1, and begin service by June 1. Building new gas turbines faces a fundamental bottleneck: GE Vernova's turbine order backlog extends through 2031, with 116 gigawatts of orders already in queue. Grid connection processes present even longer delays—half of all new generation projects now take more than five years to connect, and only 13% of projects that entered connection queues between 2000 and 2020 were operational by the end of 2025.

Why it matters

This policy shift fundamentally changes the economics and timelines of AI infrastructure deployment. Rather than treating electricity as a commodity to be purchased, hyperscale AI facilities must now function as independent power producers or accept operational risk that could interrupt training runs and inference workloads. The March 2027 proof deadline—three months before service begins—is the binding constraint, not the June service date most project teams focus on.

The approach also signals how grid operators nationwide will likely respond to surging AI power demand. Texas froze its connection queue in early August and ordered an audit of 250 to 300 projects seeking 200 gigawatts—more than double the state's peak electricity usage. Virginia introduced data center taxes driven by similar capacity concerns.

Forbes contributor Robert J. Szczerba reported these developments, noting that PJM's framework transforms firm power access from a utility service into something customers must "bring with you."