In May 2026, the North American Electric Reliability Corporation took the unusual step of issuing a Level 3 Essential Action alert, its most urgent reliability warning, over a problem that barely existed five years ago: AI data centers large enough to power small cities are connecting to the grid at a pace transmission operators cannot safely absorb, and some have dropped offline without warning, sending destabilizing voltage swings rippling through regional power systems.
The alert lands at a moment when federal regulators are already racing to catch up. The Federal Energy Regulatory Commission has opened a formal rulemaking, Docket RM26-4-000, titled “Interconnection of Large Loads to the Interstate Transmission System,” to set new rules for how these massive facilities plug into the high-voltage network and who pays for the upgrades they require. The proceeding is the clearest signal yet that Washington views AI-driven electricity demand as a direct threat to grid reliability.
The scale of the problem
NERC’s 2024 Long-Term Reliability Assessment projected that total U.S. electricity demand could grow by 38 gigawatts over the next decade, with data centers accounting for a significant share of that increase. Updated projections from grid planners suggest the data center slice alone could multiply several times over within four years, driven by the compute-hungry training runs and inference workloads behind large AI models. For context, a single hyperscale AI data center campus can draw 500 megawatts or more, roughly the output of a natural gas power plant.
The growth is concentrated in a handful of regions. Northern Virginia’s “Data Center Alley” already hosts the densest cluster of facilities in the world. Texas, Ohio, and parts of the Midwest are seeing a surge of new applications. Dominion Energy, which serves much of Virginia, has publicly stated that data center load requests in its service territory have outpaced anything in the utility’s planning history. PJM Interconnection, the grid operator covering 13 states and Washington, D.C., reported a queue of more than 100 gigawatts in pending interconnection requests as of early 2025, a backlog driven heavily by data center and clean energy projects.
Why sudden disconnections matter
The specific danger NERC flagged is not just that data centers consume enormous amounts of power. It is that they can vanish from the grid almost instantly. When a 300- or 500-megawatt facility trips offline, whether from a cooling failure, a software fault, or a deliberate shutdown, the transmission system must absorb that sudden imbalance. Generators elsewhere on the network have to ramp up within seconds to fill the gap, or frequency and voltage drop across a wide area.
Grid operators have managed large industrial loads for decades, but traditional factories and smelters ramp up and down gradually. AI data centers behave differently. Their computational loads can swing sharply as training jobs start and stop, and a full facility trip creates the kind of abrupt shock the transmission system was not engineered to handle at this scale or frequency.
The risk is not theoretical. The Department of Energy’s detailed record of the August 2003 blackout shows how a handful of transmission line failures in Ohio cascaded within hours across eight states and parts of Canada, leaving 55 million people without power. That disaster began with relatively modest triggering events. A single large data center tripping offline in a stressed corridor could, under the wrong conditions, set off a similar chain reaction.
What FERC’s rulemaking will decide
The core questions in Docket RM26-4-000 come down to money and timing. Should data center developers be required to fund transmission upgrades before they energize? Should they install on-site generation or battery storage as a buffer against sudden disconnections? Or should the costs be spread across all ratepayers, including households and small businesses that derive no direct benefit from AI workloads?
Comments filed in the docket reveal sharp divisions. Utilities and consumer advocates have argued that data center operators should bear the full cost of grid reinforcements their facilities necessitate. Tech companies and their trade groups counter that spreading costs encourages faster buildout of infrastructure that benefits the broader economy. State regulators, particularly in Virginia and Texas, have weighed in with concerns that their residents could end up subsidizing facilities owned by some of the wealthiest corporations on earth.
FERC has not set a firm deadline for a final rule. Regulatory proceedings of this complexity often stretch for months or longer, and the commission must balance speed against the legal durability of whatever standard it adopts. The risk is that data centers continue connecting under older, less stringent rules while the new framework is still being written.
What utility customers should watch
For households and businesses, the practical implications depend heavily on geography. Customers served by utilities in regions with dense data center development, especially those within PJM’s footprint, the Electric Reliability Council of Texas (ERCOT), and parts of the Midcontinent Independent System Operator (MISO), face the most direct exposure to both rate increases and reliability risk.
The single most useful step for anyone tracking this issue is to monitor FERC Docket RM26-4-000 for new filings and commission orders. State utility commission proceedings will also matter, because local regulators will ultimately decide how transmission upgrade costs flow through to retail bills.
The tension driving this story is not complicated: AI companies need enormous amounts of electricity on compressed timelines, and the grid was not built to absorb that kind of demand growth this fast. NERC’s Level 3 alert is the clearest warning yet that the gap between computational ambition and physical infrastructure is becoming dangerous. Whether FERC’s rulemaking closes that gap, or merely documents it, will shape electricity costs and blackout risk for years to come.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
