PJM Emergency Order Exposes Data Center Strain on Grid Capacity

A fall emergency on the country's largest grid

A few years ago, the second week of October was considered a quiet time for the power grid. This year it brought a federal emergency order. Energy Secretary Chris Wright invoked Section 202(c) of the Federal Power Act on Monday. His order lets PJM Interconnection call on extra generation from October 13 through October 16, and it permits backup generation as a last resort before or during an Energy Emergency Alert Level 314. PJM runs the grid across all or parts of 13 states and Washington, D.C., and says it serves more than 67 million people16. Some outlets put the number at "over 60 million," which is a rounder version of the same figure13.

The Department of Energy gave three reasons for the order: higher demand, a large share of generators down for planned autumn maintenance, and many planned transmission outages16. The margins were thin. According to the order, PJM expected demand of about 111,500 MW on October 13. It also expected roughly 61,591 MW of generation to be offline the next day, which would leave it within 900 MW of its reserve margin16. The National Weather Service forecast record or near-record warmth for parts of the Ohio Valley and Mid-Atlantic, with highs in the 80s around Washington and Baltimore17.

PJM told the public that customers did not need to do anything and that it did not expect households to be affected1611. The most likely result is a quiet week. Still, the details of the order show a lot about where U.S. electricity demand is heading.

The data center clause

The order is notable for how directly it deals with data centers. PJM said it may call on generators beyond what their operating permits allow. It may also "direct data centers and other Large Loads onto backup generation if necessary"12. The text describes PJM's aim of putting backup resources at data centers to work, "including, but not limited to, hyperscaler facilities," whether those resources are standby generators, batteries or other on-site equipment20. The Department of Energy estimates that more than 35 GW of unused backup generation exists nationwide15.

The order also lets plants run beyond their annual limits on nitrogen oxides, sulfur dioxide, particulate matter and other pollutants. Emissions during the emergency do not count against those yearly caps13.

The Trump administration has already tried to get technology companies to supply more of their own power. In February, the president signed a non-binding ratepayer protection pledge under which tech and data center companies would provide their own electricity15. This week's order goes further and treats data center backup equipment as a grid resource the operator can call on in an emergency. It is reasonable to read that as the start of something: the AI buildout's diesel generators and battery banks being folded into public reliability planning.

A regulatory collision in the same week

While DOE was widening PJM's emergency tools, the Federal Energy Regulatory Commission turned down a related, longer-term approach. PJM had proposed an Interim Resource Adequacy Service. Under it, new loads of 50 MW or more could connect without bringing matching new generation if they accepted possible power cuts during grid stress. FERC found that key parts of the plan appeared to intrude on state authority over retail electricity service12. PJM said the proposal rested on the idea that existing customers should not pay for an imbalance in which demand grows faster than new supply. It said it is reviewing the ruling to see how to meet FERC's requirements while respecting the role FERC assigned to the states12.

The result is an odd split. For four days, a federal emergency order can push data centers onto backup power. A permanent version of the same idea, letting large new loads connect on a curtailable basis, has been blocked over jurisdiction. Until that is settled, emergency orders will probably keep filling the gap.

Why PJM is short

The October event is the latest sign of a longer squeeze. In July, PJM's base capacity auction for the 2028/29 delivery year procured about 149,182 MW once fixed-resource capacity is included. That was 6,831 MW below the reliability requirement1. Industry reporting says the shortfall came mostly from rising data center load forecasts without matching new supply, and that PJM expects up to 70 GW of new large load by 20382. The July auction cleared at the price cap for the third time3. The December 2025 auction for 2027/28 had already been the first in PJM's history to miss its reliability target, falling about 6.6 GW short10.

PJM's response was a one-time backstop auction for new capacity, proposed to run from September 30 to October 212. That window overlaps with this week's emergency. The PJM board also said future auction forecasts should leave out new large loads that do not bring their own supply4.

PJM's independent market monitor has been the strongest critic. Its latest State of the Market report says current capacity is not adequate to serve large data center loads and will not be adequate in the foreseeable future. It calls the continued connection of loads that cannot be served reliably an unreasonable path that leads to more shortfalls and reliability problems6. The monitor told Utility Dive that keeping data center load in capacity auctions had raised capacity costs by $29.4 billion over the last four auctions4.

Where the accounts diverge

The reporting differs on three main points.

The cause. Wright blamed what he called past leaders' "energy subtraction policies" that left the grid exposed "when the sun doesn't shine or the wind doesn't blow"13. A DOE spokesperson went further, claiming blackouts could have risen 100-fold by 2030 without the change in policy11. PJM's explanation leaves out partisan framing. It points to demand growth, maintenance outages ahead of winter, and transmission construction that requires nearby lines to be shut down for safety16. PJM also said unpredictable weather leaves fewer days for maintenance, so autumn now carries risks it used to face only in summer and winter12. On the actual mechanics of this week, PJM's account fits the numbers better. The roughly 61.6 GW of planned outages was a scheduling and supply problem, and it had nothing to do with wind or solar16.

How unusual the order is. The Daily Mail notes that Section 202(c) orders are meant as a last resort to avoid controlled outages. DOE issued about 20 of them between 2000 and 2024, and the current administration has issued more than 60 in two years13. PJM itself said DOE had granted similar orders several times this year16, including during a July heat wave when demand neared record levels14. Coverage ranges from alarm19 to near-routine framing. The accurate reading is that the orders have become routine, and that is the warning sign. A tool meant for rare emergencies is now a standard part of operating the grid.

How large the AI demand really is. This disagreement has the most at stake, and it gets the least attention. A September 25 filing by the market monitor said 4,016.9 MW of forecast data center load had been canceled or delayed as of September 21. That is 42% of the data center growth PJM forecast between the 2026/27 and 2028/29 delivery years5. With that load removed, the monitor calculated that the 6,831 MW shortfall would shrink to 3,055 MW, a 55% reduction5. About half of the flagged load was in Dominion's zone and 29% in AEP Ohio's5. Other reporting cites projections that data center demand in PJM could triple to 29 GW by 203017. Both can be true: the long-run trend is steep while individual projects stall. But ratepayers are paying for capacity bought against forecasts that may be inflated.

What it means

Politically, the timing is bad for the AI industry. Residents and lawmakers in both parties are increasingly worried about electricity costs, strain on the grid and water use. Data center restrictions are on some state and local ballots in the 2026 midterms12. Virginia is one of the affected states and is often described as the world's data center capital18. NERC has warned that emergency federal action is only a short-term fix11.

The administration is also funding infrastructure outside PJM. A roughly $581 million package for 84 Tennessee Valley Authority projects, including $250 million in federal money, is expected to add about 2.5 GW of grid capacity14. That is far smaller than PJM's 6.8 GW gap and its projected 70 GW of new large load.

This week showed that the country's largest grid has so little slack that a warm October and normal maintenance can lead to a federal emergency declaration, and that the response is increasingly to call on data centers' own generators. Whether that becomes a lasting arrangement, in which hyperscalers act as backup power for the grid, depends on FERC, the states and the backstop auction that closes October 21. Until those questions are settled, more Section 202(c) orders are likely.