Summary
- A transmission outage in northern Virginia reportedly led large data centre loads to disconnect and switch to backup power.
- More than 3GW of load was reported offline, with PJM taking around 10 minutes to stabilise the system.
- Dense European data centre clusters may need closer coordination between facility resilience and grid operation.
PJM Interconnection experienced a voltage disturbance after a large share of data centre load in northern Virginia disconnected from the grid and switched to backup power following a transmission outage.
The disturbance was felt across a wide area, and more than 3GW of load was reported offline. PJM said the system registered a sudden drop in demand and a shift in grid frequency, while wider grid reliability was maintained.
Northern Virginia is the world’s largest data centre cluster, which gives the event significance beyond the local transmission fault. It shows how large digital infrastructure loads can affect the grid not only through rising demand, but through the way facilities disconnect, transfer to backup power, and return to normal operation.
Facility resilience can move grid risk
Data centre resilience is usually described from inside the site boundary. The facility must keep IT load running, protect customer equipment, preserve uptime, and transfer to UPS, generators, or other backup systems when supply conditions move outside tolerance. From the grid operator’s side, the same protective response can look like a sudden industrial load event.
When several facilities in a dense cluster respond to the same fault in similar ways, the effect can become large enough to affect voltage and frequency management. There does not need to be a centrally coordinated action. Protection settings, transfer sequences, UPS behaviour, and generator start logic can create a common response across separate sites.
That is where dense data centre regions become more complex for grid operators. A single large load can be studied, modelled, and managed. A cluster of large loads, each designed to protect its own uptime, creates more difficult system behaviour during faults. Sudden load loss can be as operationally awkward as peak demand if it occurs at scale.
The PJM incident should be read alongside the rapid growth of data centre load in European markets. London, Dublin, Frankfurt, Amsterdam, Paris, Milan, Madrid, and Nordic capacity clusters all face different grid conditions, but the underlying issue is shared. Large data centres are no longer passive consumers at the edge of the power system.
Backup systems need wider coordination
Backup power has traditionally been treated as a facility-level resilience asset. It protects the data hall during grid interruptions and supports service-level commitments. As facilities become larger and cluster more densely, the behaviour of those systems becomes part of grid planning.
UPS systems, diesel generators, gas generation, batteries, static transfer switches, relays, and control systems all influence how a data centre disconnects and reconnects. Their settings can affect the speed of load loss, the timing of reconnection, and the stability of the grid around the site. Those details are rarely visible in headline capacity discussions, but they can decide how a cluster behaves under stress.
Europe’s regulators and network operators may need to look more closely at dynamic behaviour in connection studies. Traditional assessments focus on whether capacity exists, what reinforcement is needed, and how demand will grow. More attention may be needed on simultaneous tripping, staged reconnection, telemetry, fault ride-through, battery support, and communication protocols between operators and control rooms.
Data centres could become part of the solution if their flexibility is designed properly. Batteries, staged load return, demand management, and grid-support services can reduce system stress. Those capabilities require commercial incentives, technical standards, and operational trust between the facility and the network operator.
The PJM disturbance does not mean data centre clusters are inherently unstable. It shows that their electrical behaviour needs to be understood with the same seriousness as their energy demand. As AI campuses and dense colocation markets grow, uptime engineering and grid engineering will have to move closer together.
