PJM Interconnection filed a proposal with federal regulators on August 13 that could rewrite the economics of every new data center built in the eastern United States. The short version: if you want to plug 50 megawatts or more into the grid, you had better bring your own power — or be first in line for the cuts when supply runs short.

The filing landed at the Federal Energy Regulatory Commission (FERC) as part of what PJM calls a multipronged response to a demand surge that is reshaping power markets across its 13-state footprint. The numbers explain the urgency. Of the 32 gigawatts of forecast electricity demand growth between 2024 and 2030 across the region, 30 gigawatts — more than 90 percent — is attributed to data centers.

That is not a rounding error. It is a structural shift in who the grid serves and who pays for it.

What PJM Is Actually Proposing

The centerpiece is a new Interim Resource Adequacy Service, or IRAS. Under the plan, new large loads that do not bring their own generation can still connect to the grid — but they enter with a condition attached. When electricity supply approaches dangerously low levels, utilities would be directed to reduce or transfer demand from those new loads ahead of any action that would shut off traditional consumers, including residential customers.

In plain English: the hyperscalers can build, but the unsupported portion of their demand sits at the back of the reliability line. PJM explicitly frames IRAS as protection for everyone else. The filing states that the unprecedented addition of large loads has given rise to resource adequacy shortfalls and real-time operational issues that the service is intended to address.

There is a second layer. Load-serving entities that take on new large loads must procure new capacity equal to or greater than the load's registered peak demand. If they fail, their zone gets directed to reduce demand during emergencies in proportion to its share of unsupported new load — and that curtailment comes before PJM calls on paid Load Management customers, who get compensated in advance for agreeing to cut usage.

The 50-Megawatt Line in the Sand

Definitions matter here, and PJM drew them sharply. A Large Load is any end-use customer with a cumulative peak of at least 50 megawatts at a single electrical site. Affiliated facilities within a one-mile radius count as a single site — closing the obvious loophole of a campus splitting into a dozen 10-megawatt buildings to dodge the threshold.

A New Large Load is one that enters service or adds incremental demand after June 1, 2027. That date is deliberate: it aligns with the Reliability Backstop Procurement process designed to fill forecast capacity shortfalls starting that same month, and it gives the market a clean line between existing commitments and everything that comes after.

Only the portion of demand not covered by qualifying new capacity is exposed to IRAS reductions. A facility that brings enough capacity to cover its registered peak simply does not face the service. The rule is not anti-data-center; it is anti-free-rider.

Bring Your Own New Capacity: The Escape Hatch

The mechanism for getting out from under IRAS is called Bring Your Own New Capacity, or BYONC — and the acronym tells you exactly how PJM thinks about this problem. Qualifying resources include new generation, certain generation uprates, surplus interconnection service, repowered resources, fuel conversions, and storage. Certain demand resources and distributed energy resource aggregations can also qualify.

Allocated Reliability Backstop Procurement capacity can cover some or all of a new load's requirement, and a load can combine BYONC, RBP capacity, and IRAS exposure to balance its position. The message is consistent: build it, bring it, or buy it — but do not expect the rest of the region's ratepayers to carry it.

That is the Ratepayer Protection Pledge in action. Introduced in March 2026 and expanded in July, the pledge says new large loads will build, bring, or buy the new generation resources needed to satisfy their energy demands, paying the full cost of those resources. PJM has also proposed that beginning with the 2029/2030 capacity auction, new large loads that do not bring their own supply simply will not be included when calculating how much capacity the region needs to procure. The load would still exist; it just would not be counted as the region's problem.

The Fight Over the Registry

None of this works without knowing who the big loads actually are, which is why the proposal creates a Large Load Registry. Schedule 11 would collect location, peak demand, ramp schedules, telemetry specifications, BYONC and RBP capacity, backup generation, and contracts with load-serving entities — information shared with states, utilities, and regulators.

That is where the first cracks are showing. The Independent Market Monitor, Monitoring Analytics, is pushing back. Executive director Joseph Bowring argues the tariff does not require data centers themselves to provide supporting evidence for the information submitted about their loads, and that the registry needs a specific validation schedule — updates at least monthly, or whenever a load's status changes, with ongoing validation and supporting evidence.

Bowring also wants explicit access for the Market Monitor rather than a vague may-be-available clause, and he points to a real-world stress test: more than 3 gigawatts of Northern Virginia data center load transferred to backup power after a transmission-line fault, which is precisely the kind of event telemetry requirements need to capture. Persistence Analytics Group has separately pressed FERC for stronger evidence requirements behind registry entries. PJM spokesman Jeffrey Shields counters that vetting has been enhanced — utilities now provide contract status, and only firm loads with an Electric Service Obligation or Construction Commitment are counted for capacity-market purposes — and that PJM is working to add state commission review and an independent third-party review of its data center forecast. The Market Monitor says it has seen no evidence of the improvement.

What This Means: The Bill Is Getting Itemized

Strip away the acronyms and this is a pricing decision. For two decades, the American grid absorbed new demand and socialized the cost of building for it. The AI buildout blew past that model. When 30 of 32 gigawatts of projected growth is one industry, the old cost-spreading logic stops working — politically, operationally, and financially.

PJM's answer is to itemize the bill. Bring your own capacity and you are a normal customer. Do not, and you are a conditional customer with a shorter fuse in an emergency. That flips the risk calculus for every developer doing site selection in PJM territory: the cost of capacity is no longer an abstract line item in the rate base, it is a direct negotiation between the developer and its load-serving entity.

The design is also careful about jurisdiction. PJM leaves retail load-reduction plans and retail cost allocation to state-level entities, which is how a regional tariff survives contact with 13 state regulatory regimes. The White House has been warning PJM to reform its governance before it is too late; this filing is the operational half of that answer.

What Comes Next

The clock is running. Comments on the filing are due September 3, 2026. PJM wants FERC to accept the petition within 60 days and to make the tariff revisions effective October 12, 2026. Meanwhile, the Reliability Backstop Procurement is slated to run in September and October, and the 2029/2030 capacity auction — the first one where unsupported new loads simply disappear from the procurement calculation — is the real test of whether this framework changes behavior.

For operators and founders running infrastructure in PJM territory, the takeaway is straightforward: power is now a build decision, not a bill you pay later. If you are planning a 50-megawatt-plus facility, the question is no longer whether the grid can host you. It is what you are bringing with you.

— Allan Ali, Sylt.ing