Fallen Power Line Exposes AI Grid Risk Near DC
28 Jul 2026
A single downed power line outside Washington, DC set off a chain reaction this week that offers a preview of a problem growing alongside the AI boom: what happens when gigawatts of data center load vanish from the grid at once.
What happened
When the power line failed, Ting Labs — a startup that runs an IoT sensor network plugged into people's electrical sockets — detected voltage spikes rippling across the PJM grid, from Northern Virginia all the way to Chicago. Lights flickered across the region, though the event stopped short of a blackout.
The trigger was a cascade of data centers switching to backup power. More than 3 gigawatts of data center load — about 3.1 gigawatts by one measurement — disconnected from the grid in roughly 30 seconds. At its peak, PJM's grid had an extra 3.49 gigawatts of electricity suddenly on it, a surplus equal to around 3% of total demand across the system at that moment. Sources differ on exactly how long it took the grid to stabilize afterward, with figures ranging from "more than 10 minutes" to 11 minutes.
PJM Interconnection, which manages grids from New Jersey to Illinois and serves 67 million customers, absorbed the shock without a wider outage. But the region where it happened matters: Northern Virginia, part of PJM's territory, has the highest concentration of data centers in the world.
Not the first time
This week's disconnection was twice as large as a similar event in 2024, when 60 data centers simultaneously dropped 1.5 gigawatts of load. That earlier episode was itself notable enough to draw attention from grid operators — and now it looks less like an anomaly and more like a pattern.
The timing is significant. Data centers made up about 6% of PJM's load in 2024; that share is projected to reach 24% by 2040. If large-load disconnection events are already happening at increasing scale, that trajectory suggests the grid could face more frequent and more severe versions of this same disruption as AI infrastructure expands.
Why the backup systems are part of the problem
Ironically, the very systems meant to protect data centers during outages may be amplifying grid instability. When data centers switch to backup power, they pull their load off the grid nearly instantaneously — which is exactly what happened this week. Ali Zain Banatwala of the Independent Electricity System Operator noted that grid operators currently lack a way for co-located loads to sequentially disconnect or reconnect, meaning many facilities in the same area can drop off in unison rather than staggering the impact.
Ricardo de Azevedo, CTO at ON.Energy, put it plainly: large load events involving data centers are "happening more and more."
Early responses taking shape
A few responses are already emerging:
- ON.Energy has built an uninterruptible power supply designed to cover entire data center campuses — servers, chillers, and other equipment — rather than just individual racks. The company is currently installing 3 gigawatts worth of these systems across four data center campuses.
- ERCOT, the grid operator for most of Texas, will require large loads like data centers to "ride through" grid disruptions rather than disconnecting abruptly. It's not stated whether PJM plans to adopt a similar requirement, or when ERCOT's rule takes effect.
- Ting Labs' sensor network, built from consumer electrical sockets, demonstrated it can detect grid-wide voltage anomalies in near real time — a capability that could feed into broader monitoring efforts.
What caused the original power line failure — weather, equipment failure, or something else — is not stated, nor is it clear which specific data centers or companies were involved. The financial or operational cost of the incident to affected parties also hasn't been disclosed.
Why founders should care
For founders building in energy, infrastructure, or AI-adjacent hardware, this event points to a few probable trends worth watching:
- Grid-resilience demand is likely to grow. With data center load share projected to quadruple by 2040, similar disruptions could become more common, suggesting rising demand for products that manage or buffer large-load fluctuations.
- Regulatory momentum may spread. ERCOT's ride-through mandate could plausibly foreshadow similar rules from other grid operators, which would create compliance-driven demand for load-management technology — though it's uncertain whether or when PJM would follow.
- Coordination gaps look like a real opportunity. The absence of protocols for sequential disconnection/reconnection among co-located data centers is an explicitly identified gap, suggesting room for software or hardware that manages load sequencing across facilities.
- Novel sensing approaches may have a market. Ting Labs' consumer-socket sensor network detecting a grid-wide event hints that alternative, low-cost monitoring architectures could carve out a niche alongside traditional grid infrastructure.
- Early movers already have traction. ON.Energy's ongoing 3-gigawatt rollout across four campuses suggests some validation for campus-scale UPS products, which could inform how newer entrants position competing or complementary solutions.
None of this guarantees a specific market outcome, but the recurrence of large-scale disconnection events — now twice as large as just two years ago — signals that grid strain from AI infrastructure growth is becoming a recurring, rather than one-off, phenomenon.