PJM's latest capacity auction cleared at $16.4 billion for the 2028/2029 delivery year, and $6.3 billion of that bill, 38 percent, traces directly to data center load, according to Joseph Bowring, president of Monitoring Analytics, PJM's independent market monitor, in Utility Dive's reporting. Prices hit the region's $325-per-megawatt-day price cap. PJM still landed 6.8 gigawatts short of its 20 percent reserve margin target, a shortfall that grew from 6.5 gigawatts the auction before.
Bowring's office has tracked data center load across four straight base auctions now. The three before this one, covering 2025/2026 through 2027/2028, added more than $23.1 billion to capacity costs from data center demand alone. Add this auction's $6.3 billion and the running total hits $29.4 billion, nearly half of the $63.6 billion PJM customers have paid across that stretch. Strip the price cap and this single auction alone would have cleared near $555 per megawatt-day system-wide, $777 in the ComEd zone covering northern Illinois, for a $29.7 billion bill instead of $16.4 billion. The price cap held this auction's cost down without solving the grid capacity shortfall driving the auction.
Bowring's diagnosis is blunt. "PJM is continuing to act like it's business as usual," he said. "You have to open your eyes and recognize that it is really a paradigm shift, and failing to do that imposes costs on other customers." His fix pulls data center load out of the shared capacity market and runs a dedicated auction where hyperscalers bid against each other instead of against every homeowner and small manufacturer on the grid. "There's only one way to do what hyperscalers agree is the right thing to do, and that is to run a separate auction," Bowring said. "That's good for the hyperscalers, because it allows them to get capacity and be served reliably." PJM's board is filing a backstop procurement plan with FERC this month, targeting September, built on a bring-your-own-generation model some hyperscalers are already running through 57 off-grid gas plants totaling 73,000 megawatts. New York regulators are fighting the same cost-allocation question without the auction mechanics, debating who pays for the grid upgrades data centers require.
Every megawatt in this fight is a nameplate number, and nameplate draw at a hyperscale campus is set by cooling architecture as much as by GPU count. A hall running air-cooled chillers at a PUE near 1.4 needs roughly 40 percent more contracted generation than the compute load alone requires. A hall on direct-to-chip liquid cooling running closer to 1.1 needs far less. Inside a shared capacity auction, that gap gets averaged away across millions of ratepayers. Inside a 15-year bilateral power contract, the kind Bowring is pushing hyperscalers toward, it becomes a number someone has to put in writing.
That is the shift buried in Bowring's numbers. Cooling efficiency stops being a facilities line item and becomes a contracted capacity commitment, priced for 15 years, signed before a single rack ships. Operators still speccing air-cooled halls for the next PJM interconnection queue are about to find out what that number costs once nobody else is left to average it against.