Who Pays for the Infrastructure?
Q1 and Q2 established what PJM needs to stay reliable: roughly 1 GW of new firm generation per 1 GW of data center load. Q3 asks what that costs consumers. We modeled three investment packages — gas-only, balanced (gas + wind + solar + storage), and nuclear (SMR) — at three data center load levels, under two cost-allocation scenarios.
Gas-only is cheap for ratepayers (1–3% increase) because gas CCGT is the lowest-cost dispatchable generation. The balanced and nuclear packages cost 3–7x more because renewables, storage, and SMRs are more capital-intensive. But the cost-allocation policy is the bigger lever: the same $140B balanced package at 50 GW costs either $10/MWh spread across everyone or $29/MWh charged to data centers alone.
The cheapest-for-ratepayers choice and the safest-for-the-grid choice are different — gas costs less but balanced is more resilient. They only align when data centers pay their own infrastructure costs.
Model: annualized CAPEX + O&M (NREL ATB 2024 costs). Gas $1.1B/GW, wind $1.3B/GW, solar $0.9B/GW, storage $300/kWh (4h). PJM total load 815 TWh/yr.
The Alternative That’s 10x Cheaper
Instead of building 20 GW of gas plants to cover peak deficits, what if data centers simply curtailed load during those hours? We ran the 2024 dispatch model and identified 109 deficit hours where the grid needs more than it has. Then we priced the curtailment across all 9 weather years (2016–2024).
Curtailment is cheaper than gas in every single weather year tested. Even in the worst year (2016, with 238 deficit hours), curtailment at $500/MWh costs $0.67B — still 5x cheaper than the gas alternative. In mild years like 2023, the ratio drops to 50x cheaper.
Why $500/MWh? That’s a generous premium — roughly 5x the average wholesale price. At $250/MWh it’s $0.15B/year (20x cheaper than gas). Even at $1,000/MWh — a penalty rate — it’s $0.6B/year and still 5x cheaper. The economics hold across any reasonable curtailment price.
Model: hourly dispatch, 30 GW DC load, 9 weather years (2016–2024). Curtailment = deficit hours × avg deficit GW × price/MWh. Gas annualized: $40B CAPEX × 8.05% CRF + O&M.
The Combined Policy Package
Q1 showed the grid needs firm capacity. Q2 showed coal retirements are the biggest risk. The interruptibility analysis above showed that curtailment beats new construction on cost. The package combines all three: keep existing coal running while you build, pay data centers to flex during the worst hours, and make sure new DC generation stays on the grid.
Coal Moratorium
Delay 10 GW of planned coal retirements. Keeps fuel-secure capacity online while alternatives are built. Reduces unserved hours from 199 to 29.
Interruptibility Program
Pay data centers $500/MWh to curtail 18% of load during peak deficit hours (109 hrs/yr). Covers the gap the coal moratorium leaves.
Grid-Connected PPAs
Mandate DC self-generation connects through the grid, not behind-the-meter. Preserves dispatch visibility and backup capacity for all ratepayers.
Coal moratorium ($2.3B) + interruptibility ($0.3B) + grid-connected PPAs ($0)
The coal moratorium buys time, not absolution. Keeping coal plants running isn’t a long-term solution — it’s a bridge while cleaner alternatives are built. The interruptibility program is the key lever: it converts data center flexibility into grid reliability at a fraction of the cost of new generation.
Coal moratorium cost: estimated O&M + fuel for 10 GW coal fleet. Interruptibility: 605 GWh deficit × $500/MWh. Grid-connected PPAs: regulatory mandate, no direct cost.