Northern Virginia Lost 3,800 Megawatts of Data Centers in 91 Seconds. The Grid's Own Watchdog Just Voted to Make Sure That Gets Fixed.
A mechanical failure near Dominion's lines triggered PJM's biggest load swing ever. Fourteen weeks later, data centers over 50MW are headed toward their first mandatory grid-reliability rules.
At 7:56 a.m. on July 22nd, a mechanical failure knocked a 230-kilovolt transmission line out of service in Dominion Energy's territory in Northern Virginia. What happened in the next 91 seconds was the largest load swing PJM Interconnection has ever recorded: roughly 3,800 megawatts of data center load, spooked by the voltage disturbance, switched itself over to backup generators almost all at once. A second wave of another 1,099 megawatts followed soon after, triggered by the first wave's own aftershocks. PJM's system-wide load dropped from just under 100,000 megawatts to about 96,200 in the space of a breath. Grid frequency ticked up to 60.092 hertz. Nine minutes later, the numbers were back to normal.
Nobody lost power. PJM and Dominion both say the bulk power system was never actually at risk. But 3,800 megawatts is roughly four large nuclear reactors' worth of load, and it vanished from the grid the way a server room goes dark when someone trips a breaker -- not because anyone wanted it to, but because the equipment inside those buildings is built to protect itself first and ask questions never. This wasn't a one-off. A similar event hit the same region in July 2024, when about 1,500 megawatts across 60 data centers dropped off after a lightning arrestor failed. Another one, also around 1,500 megawatts, happened again in 2025. Three strikes in three years, each one bigger than the last, all in the same stretch of grid that happens to host the highest concentration of data centers on the planet.
That pattern is why, on October 15th, something unusual happened: the people who actually run America's power grid voted, by a wide margin, to bind themselves to new rules most of them will have to follow.
A watchdog that mostly asks nicely just drafted a rule with teeth
The North American Electric Reliability Corporation, NERC, is the industry's own standards body -- utilities, grid operators, and increasingly data center owners sit on its committees and vote on its rules. Historically, it has had limited leverage over a facility that isn't a power plant or a transmission line. A data center drawing hundreds of megawatts from the grid has not, until now, had to prove it can ride out a voltage dip instead of running for the exits.
That changed because the Federal Energy Regulatory Commission told NERC to make it change. On July 16th, less than a week before the Dominion event but squarely aimed at the pattern building up to it, FERC issued an order in Docket No. RD26-7-000 directing NERC to file one or more mandatory reliability standards covering what regulators are now calling "computational loads" -- a category built wide enough to catch AI data centers and cryptocurrency mining facilities alike. The deadline: December 31, 2026. FERC also told NERC to update its registration rules so these facilities count as accountable entities under federal reliability law, not bystanders.
NERC had already been worried. On May 4th, it issued a Level 3 alert -- its highest non-binding warning level -- telling large computational load operators they had until August 3rd to respond to a growing pattern of sudden, self-inflicted load loss. Thomas Sisto, co-founder of the flow-battery company XL Batteries, put the scale of the problem plainly around that time: "We've seen events where 1,000MW or more dropped off the grid in a matter of seconds." Andrew Webber, CEO of the data center operator Digital Power Optimization, was more cautious about how fast a fix could actually arrive, saying it would take years of coordinated effort to get new regulations drafted and adopted. He wasn't wrong about the years. He may have undersold how fast the drafting itself would move.
What the new rules actually require
NERC's drafting committee, working under the project name Computational Load Integration, put a proposal out for an initial stakeholder ballot that passed in September. A revised version went to final ballot on October 15th and cleared it by a wide margin, well past the two-thirds threshold NERC requires. The package now moves to NERC's Board of Trustees for adoption, and from there to FERC, which has to approve it before it becomes legally enforceable under Section 215 of the Federal Power Act.
The standards create two new categories of regulated entity -- Computational Load Owner and Computational Load Operator -- covering any site with at least 50 megawatts of connected load supplied at 100 kilovolts or above. Those sites will eventually have to demonstrate they can ride through brief voltage and frequency disturbances instead of disconnecting, and existing facilities get 36 months after the standard takes effect to install the disturbance-recording equipment needed to prove it. New sites get a tighter 15-month window, or their in-service date, whichever comes later. Separately, PJM isn't waiting for NERC's effective date at all: the grid operator plans to file its own ride-through rule directly with FERC in November, ahead of the national standard, because the Dominion event happened on its system and it doesn't want to wait another year for a federal backstop.
It's worth saying plainly that this is the system working close to as designed. A real, measurable, worsening risk showed up in the data. A regulator with actual enforcement authority ordered a fix on a hard deadline. The industry's own standards body built that fix through an open stakeholder process, and the people who'll have to comply with it voted for it anyway, by a wide margin, months ahead of the deadline. That is not how every regulatory process plays out, and after a year in which this blog has spent a lot of column space on redacted contracts, contested zoning hearings, and ratepayers finding out what they're on the hook for after the fact, this one is a case where the right institutions moved while there was still time to get it right instead of after something broke.
The caveats are real, and worth naming
None of that means the job is finished. The applicability threshold that just passed, 50 megawatts at 100 kilovolts, is already higher than NERC's own drafting committee originally proposed. The early version set the line at 20 megawatts and 60 kilovolts; it moved up after utilities including Consumers Energy argued the lower bar wasn't justified by the evidence and would sweep in loads too small to matter. NERC's own analysis says 50/100 still captures about 80 percent of the computational load it's trying to reach. That also means roughly a fifth of it, by the committee's own estimate, sits below the new rule entirely -- a mid-sized campus fed at 69 kilovolts or built just under the megawatt line is outside these standards no matter how it behaves during a disturbance.
Timing is the other open question. The standards aren't enforceable yet. NERC's board still has to adopt them, FERC still has to approve the filing, and the December 31st deadline governs when NERC has to submit the package, not when a data center actually has to comply. Existing facilities get three years just to install the monitoring hardware before anyone can confirm whether they meet the ride-through requirement at all. Ben Inskeep, of the consumer advocacy group Citizens Action Coalition, has warned that the underlying grid instability from data centers "could become quite severe -- to the point of creating widespread blackouts" if the industry doesn't move faster than its own regulatory clock. He's describing a real tail risk, not a hypothetical one: PJM's own numbers show the size of these events roughly doubling with each occurrence since 2024.
What this means if you're the one signing a lease
For anyone doing real diligence on a site instead of reading the press release, the Dominion event and the standards that followed it change what "the interconnection looks fine" is actually supposed to mean. A signed interconnection agreement has never told you how a facility behaves when the line feeding it hiccups -- and after three separate multi-hundred-megawatt disconnection events in three years, that gap is no longer a theoretical risk a utility's engineers quietly manage. It is now the subject of a federal order with a filing deadline, a NERC standard most of the industry just voted for, and a grid operator filing its own interim rule because it isn't willing to wait. A facility's ride-through performance, not just its megawatt allocation, is becoming something a buyer or lender should be asking to see evidence of before the standard forces the question later and more expensively. The grid didn't blink when 3,800 megawatts vanished in 91 seconds. The next time, the people writing the rules want to make sure it doesn't have to.
Evaluating a specific parcel for a 50–300 MW load? GridPermit's Preliminary Site Kill Test grades the power, zoning, water, and commercial-readiness evidence before you buy the land.
Start a Qualification ReviewSee state-by-state coverage · Read the methodology · See a sample report