PJM weighs ride-through rules after 3.8 GW of Northern Virginia data centre load dropped off the grid
When thousands of megawatts switch to backup power in seconds, the grid's problem is not too little demand but too much sudden change.
Grid operator PJM Interconnection is considering new reliability requirements for data centres after roughly 3,800 MW of demand in Northern Virginia dropped off the grid within seconds, Data Center Knowledge reported on 12 August. The trigger was a mechanical failure on 22 July 2026 that automatically took a 230 kV transmission line out of service. As voltage was disturbed, affected facilities switched from grid supply to their own onsite generation.
The transfer came in two waves. About 2,970 MW moved to backup power initially, and a further 1,099 MW followed as system voltage rose. Losing that much demand at once forced PJM operators to cut generation quickly and deploy reactive power resources to bring voltage and frequency back into range. The operator restored its area control error within nine minutes, well inside the 30-minute requirement set by NERC, the body that writes reliability standards for North American grids, and overall reliability was maintained.
PJM describes it as the largest sudden large-load transfer it has experienced. Smaller events of about 1,500 MW each occurred in July 2024 and February 2025. Matthew Wharton, PJM's manager of reliability engineering, said the operator is evaluating potential enhancements to reliability requirements for large loads, work it is carrying out alongside the utility Dominion.
Options under discussion include voltage and frequency ride-through standards, protection coordination, rules for how onsite generation behaves during disturbances, facility reconnection protocols, telemetry and event recording. The report also quotes Neil Osnato of Persistence Analytics Group and PJM Operating Committee chair Emanuel Bernabeu.
Why it matters
Grids were designed around generators that can fail suddenly, not loads that can vanish suddenly. Data centres now cluster in gigawatt blocks with similar protective settings, so they respond to a disturbance as one enormous switch. Ride-through rules already apply to wind and solar plants; extending them to large loads would treat compute campuses as grid-scale assets. Anyone designing a facility should expect interconnection agreements to dictate its behaviour during faults.
Grids were designed around generators that can fail suddenly, not loads that can vanish suddenly.
- First wave2,970 MW
- Second wave (as voltage rose)1,099 MW
- Total transferred~3,800 MW
- Previous largest events (2024, 2025)~1,500 MW
Figures: Figures reported by Data Center Knowledge, citing PJM
What you can learn from this
- Ride-through. A ride-through requirement obliges a connected device to stay online through a short voltage or frequency excursion rather than disconnecting instantly. Generators have had such rules for years, because if every plant tripped at the first wobble a small fault would cascade into a blackout. The same logic now applies to loads: a data centre that stays on the grid for a few hundred milliseconds gives operators time to clear the fault. Conservative protection settings, chosen to safeguard expensive IT equipment, are what make facilities trip early. Rules typically specify a curve of voltage against time that a facility must tolerate before it is permitted to disconnect.
- Why losing load is as dangerous as losing generation. Alternating-current grids must balance supply and demand instant by instant, and any mismatch shows up as a change in frequency and voltage. If thousands of megawatts of demand disappear, the generators that were supplying it are suddenly pushing energy into nothing, so frequency rises and voltage climbs. Operators must throttle turbines and absorb reactive power to restore balance. The direction is opposite to a generator trip, but the physics and the urgency are identical.
- The UPS and transfer path. Data centres feed critical load through uninterruptible power supplies backed by batteries, with diesel or gas generators behind them. When utility voltage falls outside a set window, the transfer switch opens and the facility rides on batteries until generators start. From the grid's perspective the load has simply gone, and battery capacity then sets how long the facility can wait before generators must take over. Widening the voltage window that triggers a transfer, or delaying it slightly, is the practical form a ride-through rule would take.
- Reactive power and voltage. Voltage on a grid is governed less by real power than by reactive power, the component of AC flow that supports the magnetic fields in motors and transformers. When load drops, reactive demand drops too, and voltage rises. Operators deploy shunt reactors, adjust generator excitation or switch capacitor banks out to pull it back. That is what deploying reactive power resources means in the event description.
- Area control error. Each balancing authority tracks how far its net interchange and frequency have drifted from schedule; that number is the area control error. NERC standards require it to be brought back within limits inside a set time after a disturbance, which is why nine minutes against a 30-minute requirement counts as a pass. The metric matters because neighbouring grids feel the imbalance as unscheduled flows across tie lines, so one operator's problem quickly becomes everyone's.
We teach this
How to use this in practice
- Draw the power path of a data centre. Sketch utility feed, transformer, switchgear, automatic transfer switch, UPS, battery, generator and IT load as boxes with arrows. Mark where a voltage sag would be detected, which device decides to disconnect from the grid, and how long the batteries must carry the load before generators are ready. Done means you can point to the single component whose settings a ride-through rule would change, and explain what happens downstream if it waits an extra second.
- Read one grid event report. Find a NERC or grid-operator disturbance report (they are published as PDFs) and read the timeline section only. Write a five-line summary in your own words: trigger, first response, second-order effect, recovery time, lesson. Then compare it with the July event described here and note one similarity and one difference, keeping the whole summary under a hundred words so the structure stays visible. Done means you can explain the difference between a fault clearing and a load transfer without looking at your notes.
- Check a UPS you can touch. If you have a desktop or rack UPS, open its management software or front panel and find the input voltage sensitivity setting. Note the current threshold, what it would take for the unit to switch to battery, and how long the battery would last at the present load. Done means you have written down one setting that, scaled up a million-fold, is exactly the parameter grid operators are debating.
- Model the balance in a spreadsheet. Build a ten-row table with generation, load and their difference, then a column that raises or lowers a frequency figure by a fixed amount per megawatt of imbalance. Remove a large block of load in one row and watch the number move, then add a row where generation is cut in response and see how quickly the figure returns. Done means a simple chart that shows why sudden load loss pushes frequency up rather than down, and why the operator's job is to act within minutes.
Sources
- 3.8 GW Load Drop Prompts Potential PJM Data Center Rules — Data Center Knowledge
Our reporting is an original summary; full coverage is at the links above.
Don't just read about it — build it.
Square 1 teaches the skills behind the headlines, with every line of your work graded by AI. Find your starting point in 3 minutes.
Get your free skill report