PJM's power shortfall puts AI data centre growth at risk

For most of its history, PJM Interconnection was the sort of infrastructure nobody outside the industry had reason to think about. As the grid operator serving 67 million people across 13 US states and Washington DC, its job was to keep the lights on across the Mid-Atlantic and Midwest, quietly and reliably.

That reputation is now under real strain, and the reason sitting at the centre of it is the same one reshaping grids from Dublin to Johannesburg: data centre demand is growing faster than anyone can build power plants to match it.

The numbers make the scale of the problem hard to dismiss as noise. PJM’s most recent base capacity auction cleared at a record $333.44 per megawatt-day, a figure Capacity has already flagged as evidence that grid tipping points are emerging well before most forecasts expected.

Independent analysis from S&P Global Market Intelligence now projects a supply gap of around 15 gigawatts by 2030, while Bloomberg reporting puts PJM’s shortfall as high as 60 gigawatts over the coming decade if new generation isn’t built at pace. Roughly 54 gigawatts of thermal capacity has already retired across the PJM footprint since 2011, with a further 24 to 58 gigawatts at risk of retirement by 2030 without a clear replacement plan, according to legal analysis of PJM’s own filings.

A gap that keeps widening

The reserve margin gap in PJM’s base auctions widened sharply between the 2026/27 and 2027/28 delivery years, and the grid operator’s own board has now put a number on future large load growth: roughly 70 gigawatts of new demand by 2038, set against only around 15 gigawatts of generation retired since 2022. That mismatch is precisely why PJM has stopped treating this as a routine planning exercise and started treating it as an emergency.

Julia Hoos, head of USA East at Aurora Energy Research, put the underlying fragility of the current response bluntly when the plan was first unveiled.

“This is a system designed to break within a year or two,” Hoos said, according to reporting by Latitude Media.

Bring your own power, or risk being cut first

PJM’s answer, filed with federal regulators over the summer, is a genuine departure from how the grid has historically treated large customers. Under the new Interim Resource Adequacy Service, any data centre or other large load of 50 megawatts or more that connects to the grid after 1 June 2027 without securing its own generation, or otherwise guaranteeing supply, will be first in line for curtailment during a grid emergency, ahead of ordinary households. A new Large Load Registry will track every qualifying site by location and capacity, giving state regulators and utilities visibility they have not previously had. Separately, PJM has opened a bilateral matchmaking process, pairing large loads directly with new generation projects, alongside a one-time procurement auction capped at $555 per megawatt-day to bring roughly 6.8 gigawatts of new capacity online.

PJM has framed the plan as a pragmatic response to a market under pressure rather than as punitive towards the sector driving the demand. Susan Buehler, PJM’s chief communications officer, described the relationship with regulators pushing for reform as constructive rather than adversarial.

“We have advanced, through FERC filings, much of what the administration has requested,” Buehler said, responding to a request for comment reported by the American Public Power Association.

Not every stakeholder agrees the process has moved fast enough, or in the right direction. When Monitoring Analytics, PJM’s independent market monitor, filed a complaint effectively proposing a moratorium on new data centre connections, the Data Center Coalition and PJM Power Providers Group pushed back hard on procedural grounds, arguing further delay served nobody’s interests.

“A distraction and a waste of time,” the two groups said in a joint filing reported by Pennsylvania Capital-Star.

Why this matters well beyond Pennsylvania

PJM’s predicament is worth watching closely for two reasons. First, it is a live test of whether “bring your own power” can genuinely work as a grid strategy at scale, rather than as an exception for a handful of well-capitalised hyperscalers. Amazon’s purchase of a nuclear-powered data centre campus tied to the Talen Energy plant, which Capacity covered last year, was an early example of exactly this pattern inside PJM’s own territory. If that becomes the default route to reliable power rather than the exception, it will reshape site selection criteria for operators far beyond the US.

Second, PJM is not acting alone, and its response is already shaping how other US grid operators think about their own large load problems. Capacity has reported that five of the six other major US grid operators have asked FERC for more time to develop their own large-load rules, effectively watching to see how PJM’s experiment plays out before committing to their own frameworks.

Whether PJM’s mix of registries, curtailment rules and emergency procurement genuinely closes its capacity gap, or merely buys a year or two of breathing room, as Hoos suggested, will likely determine how aggressively other grids in North America, and eventually in Europe and Asia, move toward similarly binding rules for large data centre connections.

Not every market participant is convinced the current approach even holds together on its own terms. LS Power, one of PJM’s largest independent generation owners, warned regulators that repeated, ad hoc changes to market rules carry their own cost.

“Impossible for market participants to have confidence in PJM’s markets,” the company argued in a filing reported by Utility Dive.

That tension, between the urgency of closing a genuine capacity gap and the risk of unsettling the investment case for the new generation the grid actually needs, is likely to define PJM’s next 18 months. For any operator planning US East Coast capacity, it is no longer a regulatory footnote. It is the central variable in whether a project gets built on schedule at all.