FERC Just Froze PJM's $20 Billion Fix for Its Power Shortfall. Amazon's Own Fix Closed the Next Morning.
PJM's emergency backstop auction for a 6,831-megawatt capacity gap was supposed to open September 30th. FERC accepted it, then suspended it for five months over who pays. The next morning, Amazon signed its own 20-year, 690-megawatt nuclear deal in the same state — a deal the frozen auction had nothing to do with.
We wrote last week about PJM's plan to close a 6,831-megawatt capacity shortfall with a one-time "Reliability Backstop Procurement": up to $20 billion authorized, a price cap of $555 per megawatt-day, bidding open from September 30th through October 21st. It didn't open. On September 29th, FERC accepted most of the design and then suspended it for five months, effective February 28th, 2027, subject to refund, over exactly the cost-allocation fight five state ratepayer advocates had flagged before the window even started. PJM pulled the auction the next morning with no new date set. Eighteen hours after that, Amazon signed its own fix — a 20-year, 690-megawatt nuclear power purchase agreement with Constellation, in the same grid, the same state, the same week the collective mechanism froze.
What FERC actually objected to
FERC's order, in Docket No. ER26-3380, didn't reject PJM's backstop outright. Chair Laura Swett and the rest of the commission accepted the core of it — the $555 price cap, the authorization to go procure up to 6,831 megawatts of new capacity on 15-year commitments. What they wouldn't accept, on the record as filed, were three pieces of the fine print that determine who actually pays for it. PJM's method for assigning backstop costs used a "snapshot" load adjustment that FERC found doesn't capture forecasted large-load growth, meaning some load-serving entities could end up paying for capacity benefiting load that shows up in their territory later, after the snapshot was taken. PJM's rule for a transmission owner that exits the grid mid-commitment would have left remaining obligations with that owner rather than the load-serving entities actually in its zone, which FERC said gets the responsibility backward. And PJM hadn't adequately justified the collateral it planned to require from the utilities buying power under the mechanism. "This commission will not be forced into accepting a deeply flawed, eleventh-hour procurement mechanism with billion-dollar implications for consumers," Swett wrote. Commissioner Judy Chang See was blunter about the stakes: "Existing customers should not be left paying costs attributable to new demand." Commissioner David Rosner, concurring separately, made the opposite case for why the commission accepted the framework at all rather than killing it: the responsibility to the 67 million people PJM serves is to keep the lights on, and the downside risk of inaction outweighed the risk of an imperfect mechanism going forward while the unresolved pieces get litigated on paper.
The deadline that didn't move
Here's the detail that matters most for anyone with a site inside PJM's footprint: FERC's suspension froze the procurement itself, but it didn't touch the opt-out clock running alongside it. Maryland's Office of People's Counsel — the same office that put the state's exposure at up to $562 million over 15 years in our last post on this — had proposed letting a utility's zone reduce or escape backstop costs if it could show its data-center customers participate in genuine demand-response programs rather than running flat out year-round. The reporting deadline for that opt-out was October 21st. It still is. Utilities now have to decide whether to file demand-flexibility data against a cost-allocation formula FERC has just told PJM, on the record, may be unjust and unreasonable and is getting rewritten in a paper hearing. That's a real decision with real consequences, made against a mechanism that is itself suspended and could come back in February looking different from what anyone is opting in or out of today.
The next morning, Amazon didn't wait
On September 30th — the day PJM's auction was supposed to open and instead didn't — Amazon and Constellation announced a 20-year power purchase agreement covering 690 megawatts at Calvert Cliffs Clean Energy Center, Maryland's only nuclear plant. Part of that capacity is the plant's existing output; roughly 190 megawatts is new, from an uprate Constellation will build out between 2030 and 2032, backed by more than $3 billion in Maryland infrastructure investment. A related retail supply agreement extends the arrangement to cover Amazon's operations across PJM's 13-state footprint. None of it ran through PJM's capacity auction, backstop or otherwise. A generator and a buyer with the balance sheet to sign a two-decade bilateral contract simply agreed on a price and a delivery date, in the same state whose ratepayer advocate had just told FERC it couldn't afford the collective version of the same problem.
The diligence angle
On our evidence ladder, these two outcomes don't sit anywhere close to each other, and the gap between them is the actual story. PJM's backstop procurement, even fully approved, would have been a regional cost-allocation mechanism — real money, but spread across every load-serving entity in the footprint according to a formula FERC just sent back for rework. Suspended, subject to refund, with three of its own load-mechanics unresolved until at least February, it's currently worth less on the ladder than the agency-confirmed rung, not more: a plan everyone agrees is needed, with the terms everyone needs to rely on still being litigated. The Amazon-Constellation deal is the opposite case. It's a signed, dated, 20-year contract with a named capacity figure and a named delivery window — rung five, contractually committed, the top of the ladder — for the single company able to write that kind of check. That's the same gap we flagged when Helix Digital Infrastructure's "preferred power provider" relationship with Vistra turned out to be real at the corporate level and untested at any specific site: a hyperscaler's access to bilateral power is not a market condition everyone else in the queue gets to rely on. GridPermit doesn't work inside PJM's footprint, but the lesson travels: when the collective fix for a capacity shortfall stalls in a cost fight, the gap doesn't close evenly. It closes first for whoever can skip the queue, and stays open longest for whoever can't — which, on most 50-to-300-megawatt projects we look at, is nearly everyone bidding against Amazon for the same constrained grid.
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