America's largest power grid told data centers: bring your own power, or get cut off first

PJM, the largest US power grid operator, is ending the era of data centers drawing freely from the shared grid. The operator that manages the wires and wholesale market for about 67 million people across 13 states and Washington, D.C. filed a two-part proposal this month at the Federal Energy Regulatory Commission: up to $20 billion in capacity payments to underwrite new power plants for data centers already in the pipeline, and a hard rule that anything built after 2027 must bring its own generation, or risk being curtailed before homes and businesses when the system is short.

Once the rule takes effect in mid-2027, new data centers that cannot feed themselves with their own generation will be served only at the back of the line. PJM would be able to curtail them before non-data-center customers, in effect reclassifying hyperscale compute as a second-tier load on the country's largest grid.

Up to $20 billion will be drawn from the same wholesale capacity market that funds other generators. PJM plans to issue a request for proposals on September 30, 2026, and award contracts in December 2026, to underwrite new power plants capable of supplying data centers expected to come online through 2027. The implicit trade: data centers already in the queue, and their host communities, get a financed bridge. The next wave pays for itself.

New supply on the PJM grid has not kept pace with the data-center boom, and the gap is the reason prices for everyone else on the grid have surged. By writing "bring your own generation" into the tariff, PJM is acknowledging that the default of letting hyperscalers land anywhere and draw from the shared system has reached its limit on the operator's 13-state plus Washington, D.C., footprint.

FERC's earlier compliance order required PJM to revise its transmission tariff to permit new services for co-located loads, the regulatory term for data centers paired with their own power plants. PJM's accompanying large-load plan layers on a price collar and an expedited generator interconnection track to make the bring-your-own option actually buildable. The August proposal packages those pieces and asks FERC, on the E-1 co-location proceeding docket, to bless the whole thing by October 12, 2026, so the post-2027 rule can take effect next summer.

A hyperscaler planning a 2028 campus in northern Virginia, central Ohio, or the PJM edge of Pennsylvania can keep its existing interconnection date only if it has, or contracts for, its own megawatt-hours. Otherwise it gets a slot in the curtailment queue. The proposal does not name specific operators or sites, and the question of which data-center projects will sign generation contracts versus which will redesign around the curtailment risk is now an active one for hyperscalers, their offtakers, and the independent power producers that have been queueing plants to serve them, as industry analysis of the underlying FERC order has laid out.

Ratepayers have already absorbed the price of a planning shortfall that left the grid short of capacity relative to demand. The proposal splits that cost: a $20 billion bridge for what is already coming, and a hard wall for what comes next. Whether the wall holds depends on FERC's order, on whether PJM's generation RFP actually delivers by mid-2027, and on whether the expedited interconnection track shrinks the multi-year queue for any new power plant, data center or otherwise.

PJM's own framing admits the price surge is its own doing: the operator has not been able to bring new supply online fast enough to keep up with data-center demand, and that is the gap the new rule is meant to close. The two-part filing is the operator's answer to a problem it helped create. Whether that answer is enough, or just the first tariff to land in a regional race to internalize AI's electricity cost, will be tested first in FERC's order and then in the queue for the next gigawatt.