PJM Wanted the Power to Decide Which Data Centers Get Cut Off First. FERC Just Said That's Not PJM's Call.
FERC approved a registry tracking every 50-megawatt-plus data center in 13 states, then rejected PJM's plan to curtail them ahead of ordinary demand response. The $29.4 billion question of who pays is still open.
Picture a bad August afternoon somewhere in PJM's 13-state footprint, the grid straining under a heat wave, and the operator needs megawatts back right now. Whose power goes first: the hospital, the data center next to it, or some household three counties over that never signed up for any of this? PJM spent months building an answer into its own rulebook. On October 9th, the Federal Energy Regulatory Commission told PJM that answer isn't PJM's to write.
What PJM actually asked for
PJM filed its plan with FERC on August 13th, built around three pieces. The first was a formal definition: a "Large Load" is 50 megawatts or more of demand at a single site, or across affiliated sites within a one-mile radius. The second was a Large Load Registry, giving PJM visibility into the location, ramp schedule, and capacity arrangements of every load that size in the 13-state, 65-million-person territory it runs. The third, and the one that mattered most, was something called the Interim Resource Adequacy Service, or IRAS: a fast lane that would let a new data center connect to the grid before the capacity needed to serve it actually existed, in exchange for PJM getting to curtail that load first, ahead of other demand, the moment things got tight. PJM also asked to exclude these new large loads from the capacity auction's demand curve starting with the 2029/30 delivery year, and to let loads that bring their own generation, a provision called "Bring Your Own New Capacity," skip the arrangement entirely.
FERC's order, in Docket ER26-3515, split the plan in half. The Large Load definition and the registry were approved and took effect October 12th. Everything built around curtailment — IRAS, the Bring-Your-Own-New-Capacity carve-out, the demand curve exclusion — was rejected. FERC's objection was jurisdictional: deciding who gets cut off first, and in what order, is a retail question that belongs to the states PJM operates in, not to a regional grid operator's own tariff.
The design that bothered even a commissioner who agreed with the goal
Commissioner David Rosner didn't dissent, but he filed a concurrence that reads like a warning label. He agrees with the underlying instinct: in an emergency, a non-critical data center should go dark before a home, a hospital, or a school. What he objected to is how PJM's plan would have gotten there. IRAS load would have been curtailed in the earliest stages of an emergency, Rosner wrote, before PJM had even called on voluntary demand response, the customers who've agreed in advance to cut back for a payment. That ordering bothered him for a reason beyond fairness to data centers: if PJM can shed thousands of megawatts of demand involuntarily while its capacity market keeps reporting the region as adequately supplied, the market stops telling anyone the truth about how tight the grid actually is.
States need to create rules for deciding which customers are cut first.Commissioner David Rosner, FERC, concurring in Docket ER26-3515
I think that's the real finding buried in this order. FERC wasn't just protecting state authority for its own sake. It looked at a specific mechanism and decided the sequence was backwards: voluntary cuts first, mandatory cuts last, and the decision about who's in which category made by the regulator that actually answers to the ratepayers involved. PJM, a grid operator with no elected accountability to anyone, had proposed writing that sequence itself.
What the registry actually buys
The part of this order that survived is not nothing. A list of every 50-megawatt-plus load in PJM's territory, with location and ramp schedule attached, is exactly the kind of agency-confirmed fact GridPermit's own evidence ladder looks for, and it's a real gap closed: before October 12th, nobody outside each individual utility had a consolidated, regulator-mandated view of how much large load was actually sitting on this grid, as opposed to the load PJM was forecasting. PJM told FERC it expects 32 gigawatts of new demand between 2024 and 2030, with 30 gigawatts of that coming from data centers. A registry won't change that number, but it means the next fight over it happens with an actual list in the room instead of competing projections.
The number nobody's attached a bill to yet
Here's what the registry and the curtailment fight are both really about. Monitoring Analytics, PJM's own independent market monitor, found that data center load added $29.4 billion to the cost of PJM's last four capacity auctions, about 46% of their combined $63.6 billion total, and called the increase irreversible. That's the backdrop against which the Natural Resources Defense Council read this ruling as a loss, not a win: with IRAS dead, NRDC argues, PJM still has no mechanism forcing new large loads to bring their own supply, which means the public keeps absorbing the capacity cost of data center growth until a state, or PJM's next filing, does something about it. No regulator, utility, or analyst has published a bill estimate tied specifically to this order. Nothing about a household's rate changes today. The $29.4 billion question just moved from PJM's tariff to thirteen separate statehouses, which is either the appropriate venue or a very long detour, depending on how much you trust your own state commission to get there before the next capacity auction clears.
PJM's own reaction has been notably unbothered in public. A spokesperson said the grid operator is reviewing the order with an eye toward fulfilling FERC's requirements while recognizing the responsibilities FERC says belong to the states — which is the sound of an organization that expected to lose part of this filing and already knows what its next one will say. FERC gave PJM 30 days to refile clean tariff language reflecting what survived. The curtailment question, the one that actually decides whose power goes out first on that bad August afternoon, goes back into the queue.
Evaluating a specific parcel for a 50–300 MW load? GridPermit's Preliminary Site Kill Test grades the power, zoning, water, and commercial-readiness evidence before you buy the land.
Start a Qualification ReviewSee state-by-state coverage · Read the methodology · See a sample report