---
title: "3,800 megawatts left the grid in seconds. No equipment failed."
url: "https://3dfs.com/articles/3800-mw-left-the-grid"
description: "On July 22, 2026, Northern Virginia data centers dropped about 3,800 megawatts of load during a fault the grid cleared normally. Why the loads left."
updated: "2026-10-15"
---

# 3,800 megawatts left the grid in seconds. No equipment failed.

> On July 22, 2026, Northern Virginia data centers dropped about 3,800 megawatts of load during a fault the grid cleared normally. Why the loads left.

On July 22, 2026, data centers in Northern Virginia disconnected about 3,800 megawatts of load during a fault the transmission system cleared normally. The generators stayed up. The wires stayed up. The loads left.

PJM’s analysis points at load-side sensitivity: protection settings inside the facilities read a survivable sag as a threat and transferred off grid. Each facility made a locally sensible choice. The choices synchronized, and the grid absorbed a 3,800-megawatt step change.

The regulatory answer is arriving as ride-through requirements. The engineering answer is to give the protection logic nothing to flee: hold voltage quality at the panel through the sag.

The full anatomy is on [our Data Centers and Compute page](https://3dfs.com/solutions/data-centers-compute).

## How a cleared fault becomes a load loss

The [Data Centers and Compute page](https://3dfs.com/solutions/data-centers-compute) carries an earlier record of the same kind. On July 10, 2024, a lightning arrester failed on a 230 kilovolt line, and reclosing produced six faults in 82 seconds. About 1,500 megawatts of data center load disconnected, and utility equipment tripped none of it. Most of the loss left on the third sag.

Both records make the same point: protection inside the facilities read a cleared fault as a threat. The fault was over. Facility schemes count sags, typically three in one minute, and transfer to backup until reset. In Texas, NOGRR282 now prohibits sag-counting trip schemes for sags the load must ride through.

The supplies behind that protection add to the problem. A regulated supply holds its output power, so its current rises as the voltage falls. The rise adds to the drop across the source impedance. The [Voltage Stability page](https://3dfs.com/technology/voltage-stability) follows the same chain from the panel: each voltage fault there has a current fault behind it.

## Ride-through belongs to stored energy

An uninterruptible power supply carries the load through a sag, and a series sag corrector restores the missing voltage. Correction at the panel works on a different share of the drop. It removes the share that the loads’ reactive, harmonic and unbalanced current makes across the plant’s own impedance. A deep sag from the utility side stays with stored energy.

## Where this stands today

The [Data Centers and Compute page](https://3dfs.com/solutions/data-centers-compute) now carries this event. It reports Dominion’s preliminary finding on the cause: a sub conductor pulled free from a compression sleeve. PJM’s preliminary finding attributes the transfer to voltage ride-through performance and states that the fault cleared properly.

The same page states a limit of our controller: it does not hold the voltage the source delivers and does not ride through an interruption. The [Standards Compliance page](https://3dfs.com/technology/standards-compliance) lists ERCOT’s rule and the FERC order for large computational loads. Start with a one-week measurement at one power distribution unit.

### Related reading

- [Resilience](https://3dfs.com/technology/resilience): what a short sag costs a plant, and what correction does not change

- [Request a site analysis](https://3dfs.com/site-analysis): the facts an engineer reads before saying what to measure first