FERC Accepts PJM's Data Center Backstop Auction, Then Suspends It Five Months Over Who Pays

PJM’s backstop auction for new capacity, the one built around data center load growth, was supposed to open its offer window today, September 30. It did not. On Tuesday the Federal Energy Regulatory Commission accepted PJM’s Reliability Backstop Procurement in Docket No. ER26-3380-000, suspended it for five months to February 28, 2027, and sent three pieces of it to a paper hearing ^[1]. PJM said the process “will not begin on Sept. 30 based on the requirements of FERC’s order,” and that “actual timing is to be determined” ^[2]. The target was 6,831 MW ^[1].

This is a federal decision on a grid operator’s tariff. States are not out of it, as the order itself goes out of its way to say.

What FERC accepted and what it set aside

The backstop, or RBP, is a one-time procurement of new capacity that PJM describes as needed because of “an influx of large loads” ^[1]. Sellers of new resources would get fixed commitments of up to 15 years, starting no earlier than the 2028/2029 delivery year and ending no later than 2042/2043 ^[1]. Committed resources would offer into PJM’s regular capacity auctions at $0/MW-day and settle against a contract for differences, so load is not charged twice for the same megawatt ^[1].

The 6,831 MW starting target is the shortfall from the 2028/2029 Base Residual Auction, which cleared in July at the temporary price cap of $325/MW-day ^[1]. The auction before it, for 2027/2028, came up 6,623 MW short, and FERC’s order notes that nearly 5,100 MW of the 5,250 MW of forecast load growth behind that shortfall was data centers ^[1].

Most of the design survived. FERC found PJM’s $555/MW-day maximum willingness to pay, measured as a MW-weighted average over the delivery period, and its pay-as-bid selection process just and reasonable ^[1]. It found excluding Fixed Resource Requirement entities just and reasonable, since those utilities already carry their own capacity outside the market ^[1]. Seller collateral passed ^[1].

Three items did not ^[1]:

• Cost allocation among PJM’s Electric Distributor zones.
• Transmission owner exit rules, which govern who pays for 15-year commitments if a utility leaves PJM.
• Collateral for load-serving entities, the buyers’ side.

FERC also found that PJM’s opt-out for qualifying cooperatives and municipal utilities “may be unjust, unreasonable, and unduly discriminatory,” because the large loads those utilities serve are “similarly situated to large loads served by other LSEs” ^[1].

The order sets the procedure. PJM files an initial brief within 45 days; responses follow 20 days after that. FERC “would grant a motion” to hold the briefing in abeyance if PJM files a new section 205 proposal within 30 days ^[1]. By this column’s count, those dates are November 13 and October 29. The order also opens a section 206 investigation, Docket No. EL26-108-000, and FERC says it expects to decide within five months ^[1].

Who pays: the snapshot problem

This is the part that matters to data center developers, so it gets the room.

PJM proposed to split the target among Electric Distributor zones by each zone’s pro rata share of the change in forecast Load Adjustments between 2026/2027 and 2028/2029 ^[1]. Load Adjustments are utility-reported changes to summer peak load at the zone level, layered onto PJM’s base forecast ^[1]. That split would hold for the whole 15-year term ^[1]. Inside each zone, states or other retail regulators would decide which customers pay, and if a distributor reported nothing 60 days before the 2028/2029 delivery year, PJM would spread the charge across all load in the zone the way it spreads ordinary capacity costs ^[1].

FERC accepted the general idea of a fixed “snapshot” charged to the zones responsible for the near-term shortfall ^[1]. What it questioned was the measuring stick. Some large loads, the order says, “may never have been accounted for as Load Adjustments” because PJM had already folded them into its base forecast, so the target “may only capture a portion of new load growth, potentially leaving out some new large loads” ^[1]. Its suggested fix: allocate on the difference between actual 2026/2027 load in each zone and the latest 2028/2029 forecast, and use current forecasts for the target too ^[1].

The record shows why the measuring stick drew fire. The New Jersey Board of Public Utilities pointed out that utilities reported Load Adjustments differently, with FirstEnergy using loads of 50 MW or more and Duke Energy reporting expansions of 20 MW or more ^[1]. PJM’s independent Market Monitor argued the 6,831 MW target was already stale, estimating that 4,017 MW of data center load had been cancelled or delayed as of September 21, which it put at 42% of the 9,481 MW increase in data center load between the 2026/2027 and 2028/2029 auctions ^[1]. Joint consumer advocates noted that PJM’s large load forecasts had fallen from 14.8 GW to 6.8 GW since April ^[1].

FERC’s closing word on cost was aimed at the states. It quoted the January principles signed by all 13 PJM governors, which commit them to have their utilities allocate backstop costs “to new data center loads that have not otherwise procured capacity or agreed to be curtailable” ^[1]. Then: “we will not pretend that all of these issues can be solved exclusively by market design in the federal sphere. The PJM states have their own role to play” ^[1].

That is the same cost question Virginia’s governor put to PJM in Executive Order 22, which, as this column noted on September 23, told the state’s Chief Energy Officer to push for assigning backstop costs to the data center loads causing them. FERC has now said, in effect, that the zone-level split is its job and the customer-level split is the states’.

Where the sides stand

PJM put the order in a positive light. “PJM welcomes FERC’s acceptance of our Reliability Backstop Procurement proposal,” it said, adding that the commission “identified three discrete issues requiring further consideration” and that PJM intends “to work quickly” on them ^[2].

Chairman Laura Swett did not. In a concurrence, she wrote that the proposal arrived “on the last possible statutory day” to open a September 30 window, and that “this Commission will not be forced into accepting a deeply flawed, eleventh-hour procurement mechanism with billion-dollar implications for consumers” ^[1]. Commissioner Lindsay See said regulators and grid operators “need to adjust to sustained and asymmetric growth,” according to Utility Dive ^[3].

The data center side was not of one mind. Amazon argued that a split fixed before the load materializes will drift away from actual cost causation, and asked for periodic updates as load actually shows up, warning that a frozen split invites new large loads to site in zones with small obligations ^[1]. The Data Center Coalition supported letting states decide the allocation to large loads ^[1]. Equinix asked FERC to recalibrate buyer collateral to foreseeable exposure ^[1].

Cooperatives split as well. Northern Virginia Electric Cooperative said its highest possible collateral requirement, about $1.95 billion, would exceed its total assets, and objected that Old Dominion Electric Cooperative could opt out while it could not ^[1]. FirstEnergy argued the cooperative opt-out was discriminatory because large load in those territories is just as real as anywhere else ^[1]. FERC sided with FirstEnergy for now.

PJM proposed an initial buyer credit rate of $1.5 million per committed megawatt, about half the contract value at the cap ^[1]. FERC said PJM had not backed the 50% figure with any quantitative measure, such as a probability of default ^[1].

What I’m watching

First, whether PJM files a new section 205 proposal by October 29, by this column’s count, and whether it adopts FERC’s actual-load allocation or defends Load Adjustments ^[1]. PJM has scheduled an administrative update for the October 7 Market Implementation Committee meeting ^[2].

Second, the target. FERC told PJM to use “the most up-to-date load and resource forecasts” for the initial number ^[1]. If the Market Monitor’s 4,017 MW is close, 6,831 MW is the wrong number. Nobody has published the right one.