The U.S. Grid Needed an Emergency Order in October. The Real Crisis Is Who Pays for Data Centers.

On Tuesday, October 14, the operator of the largest power grid in the United States was running on emergency authority. The Department of Energy had issued an emergency order, effective October 13 through October 16, authorizing PJM Interconnection to fire up extra backup generation to keep the lights on across 13 states and Washington, D.C. It was the second such order in two months. The calendar said October. The grid was behaving like August.

The immediate trigger was weather, but the deeper cause is structural. A record-breaking October heat wave collided with the fall maintenance season, when generators and transmission lines are routinely taken offline for repairs. Underneath that collision sits the slower emergency: electricity demand, driven overwhelmingly by new data centers, is growing faster than new power plants can be built. Days earlier, federal regulators had rejected PJM's plan to manage exactly that problem. The weather promptly exposed what the rejection left unresolved.

A heat wave in the middle of maintenance season

October is supposed to be the grid's easy season. Utilities schedule their heaviest maintenance between mid-September and early December precisely because demand is mild. That plan assumes the weather cooperates. It did not. The National Weather Service reported unseasonable, record-breaking warmth across the Southern Plains, the Southeast and the mid-Atlantic, with temperatures in PJM's service region topping out in the high 70s and 80s on October 14, reaching up to 90 in places, before dropping as much as 20 to 25 degrees later in the week.

The result was a grid stretched uncomfortably thin. PJM said it expected demand of 113,312 megawatts on October 14, and that the system would come within 900 megawatts of its reserve margin. The most acute stress concentrated in the territories served by Baltimore Gas and Electric, Pepco and Dominion Energy, covering parts of Washington, D.C., Maryland and Virginia. PJM said it had canceled transmission and generation outages "to the extent possible."

Energy Secretary Chris Wright said the region faced "elevated demand, a high rate of planned generator outages during the autumn maintenance season, and numerous planned transmission outages," and that the order would make additional generation, including backup power, available as needed. One provision mattered more than the rest: under the order, PJM could redirect data centers and other large loads onto backup generation. It was a dry run of exactly the authority PJM had wanted permanently, and that regulators had just denied.

The calendar said October. The grid was behaving like August.

The ruling that denied PJM its permanent fix

The emergency order landed five days after a decision that had been simmering all week. On October 9, the Federal Energy Regulatory Commission rejected the core of PJM's plan to cope with large new electricity loads, principally data centers. The rejected proposal had two main pieces. IRAS, the Interim Resource Adequacy Service, would have let PJM require certain large electricity customers to cut consumption during grid emergencies if they had not secured additional power generation. BYONC, Bring Your Own New Capacity, would have given large customers a formal route to do exactly that: bring their own new generation to the grid.

FERC's grounds were jurisdictional and economic. The commission wrote that IRAS and BYONC were either outside its authority or had not been shown to be just, reasonable and non-discriminatory. The decisive point was that deciding which retail customers get cut first is a state matter: a federal grid operator's tariff cannot unilaterally dictate curtailment priorities among retail customers. PJM spokesperson Jeff Shields said the plan was meant "to address the growing electricity reliability challenge where electricity demand is increasing faster than new generation is coming online." The commission agreed the problem was real, calling the capacity shortfall imminent, but rejected the mechanism.

FERC did not reject everything. It accepted PJM's definition of a Large Load, end-use customer demand of at least 50 megawatts at a single electrical site, with affiliated facilities within one mile counted as a single site so an operator cannot dodge the threshold by building a cluster of smaller buildings, and a Large Load Registry to track them, effective October 12. Load-serving entities must now report each large load's location, in-service dates, peak load and ramp schedule, telemetry and backup generation, while PJM publishes only aggregated data. The registry gives PJM the visibility it lacked. It does not give PJM the stick it wanted. PJM has warned it expects to fall short of sufficient capacity to meet regional reliability requirements beginning with the 2027-2028 delivery year, specifically because of data center demand and the pace at which new facilities connect.

Who pays for AI's electricity

The stakes behind the acronyms are the same ones showing up on household bills. PJM's capacity prices, the payments that keep generators available for peak hours, are pinned at the cap of $325 per megawatt-day for the 2028/29 delivery year, and investment bank Jefferies expects them to stay higher for longer. PJM expects data center demand on its system to roughly triple by 2030, reaching about 29 gigawatts. One analysis of PJM market data put the data center share of capacity costs at $29.4 billion through 2028/29, about 46 percent of the total. Those are analyst estimates from market data, not official figures, but they show why the question of who pays has become the central fight in American electricity policy.

Shields summed up PJM's argument this way: existing customers should not bear the consequences of the supply-demand imbalance, and data centers and other large loads should fulfill the commitments they made to help maintain reliability and affordability. FERC left the door ajar, inviting a revised proposal in a 30-day compliance filing, and effectively moved the retail curtailment question to the states. The decision over who gets cut first has therefore descended to 13 state capitals plus the District of Columbia, each with its own politics.

The three clocks

Analysis: the grid is now running on three clocks that do not agree. The weather clock runs in days, and it just produced a heat wave in October. The capacity clock runs in years: generators take years to build, and PJM is already warning about 2027-2028. The regulatory clock runs in months, and it just spent its time deciding who is allowed to decide. The emergency order bridges the first clock. Nothing in the past week bridged the other two. Until new generation gets built faster than new data centers connect, mid-October will keep feeling like mid-August on the American grid.