Skip to main content
PJM Data Center Study → Question 2

Can PJM Survive Another Elliott?

Winter Storm Elliott (Dec 2022) forced 46,000 MW offline — 24% of PJM's fleet. Gas outage rate hit 37% (FERC/NERC confirmed). The grid barely avoided rolling blackouts. What happens if it hits again with data centers?

Vulnerability Surface

Gas Is PJM's Achilles Heel

Hours Unserved: Gas Curtailment vs. Data Center Load
Finding
At 10% gas curtailment, the tipping point drops from 25 to 15 GW. At 30% — roughly Elliott severity — the grid can't serve current demand, let alone data centers.

88 GW of PJM's dispatchable fleet runs on gas, and gas supply is correlated with cold weather demand — when you need it most, it's most likely to fail. The vulnerability surface shows this isn't a cliff edge. It's a slope, and every GW of data center load steepens it.

Model: hourly dispatch with parametric gas curtailment (0–70%) across 9 weather years. 2022 weather includes actual Elliott demand spike.

Elliott Counterfactual

Christmas Eve, Replayed

Homes Without Power During Elliott (37% Gas, 16% Coal Outage Rates)
Key Finding
If 30 GW of data centers had existed during Winter Storm Elliott, 22 million people would have been without heat on Christmas Eve — 13 million more than without data centers.

Calibrated against FERC/NERC final report (Nov 2023). Actual gas outage rate was 37% (vs. our assumed 32%). Model understates the real risk.

Zonal Breakdown

Dominion Zone Breaks First

The Dominion zone (Northern Virginia) has 20 GW of generation serving 70% of DC load, with only 6.6 GW of transfer capacity from the rest of PJM. It fails in 9 out of 9 weather years at 10 GW DC while PJM system-wide shows zero hours unserved.

Dominion Zone vs. PJM System: Hours Unserved by DC Load

Named substations BECO, Mars, and Wishing Star carry the bulk of Northern Virginia interconnection. PJM's planned $4.8B HVDC backbone adds 3 GW of transfer capacity by 2032 — but when the zone needs 13 GW of relief at 15 GW DC load, 3 GW isn't enough.

Calibrated against: EIA-860 generator data, NRC nuclear capacities, PJM CETL transfer limits.

Cascade Analysis

The Gas-Electric Cascade

When gas plants trip, remaining units run harder, increasing their failure probability. This feedback loop amplifies outages far beyond the initial shock. The cascade model tracks this S-curve progression, calibrated against the FERC/NERC timeline showing escalation over 36 hours during Elliott.

Cascade Amplification: Hours Unserved at 30 GW DC Load
Finding
The cascade amplifies failures 2.3x at Elliott severity. A moderate 10% initial outage cascades to 1,225 hours at 30 GW DC. The S-curve means small perturbations trigger disproportionate consequences.
Fleet Trajectory

Planned Buildout Makes It Worse

Coal retirements reduce fuel-secure capacity faster than new generation comes online. The planned ~2028 fleet is less resilient than the current fleet under identical stress conditions.

Fleet Firm Capacity Elliott Hours (0 GW DC) Elliott Hours (15 GW DC) Elliott Hours (30 GW DC)
Current 2024 ~160 GW 164 714 2,342
Planned ~2028 ~153 GW 291 1,103 3,102
Full Plan ~2032 ~145 GW 472 1,628 3,927

The current fleet can't survive another Elliott even with zero data centers. PJM avoided load shedding in Dec 2022 through emergency procedures, conservation appeals, and luck. The ~2032 fleet has 472 hours of unserved energy under Elliott conditions at zero DC load — nearly 3x worse than the current fleet.

Validation

Our Assumptions Were Conservative

FERC/NERC confirmed our assumed gas outage rate was conservative: actual Elliott forced outage rates were worse than what we modeled. Gas peaked at 37% (not our assumed 32%), coal at ~16% (close to our 15%). Marcellus shale gas production dropped 23%, Utica dropped 54%. 92% of plant failures gave less than one hour's notice.

The model understates the real risk. Every finding on this page uses outage rates below what actually occurred. The true vulnerability surface is worse than shown.