#00221
Connect large loads sooner if they accept curtailment during scarcity hours. Modelling finds 76 to 126 GW of headroom in the 22 largest US balancing authorities at 0.25–1% of hours curtailed. Google has signed 1 GW of demand response but publishes no curtailment performance data.
Parent issue
#00208 Large data-center load shifts grid and capacity costs onto households and other ratepayers
Location
Description
A utility offers a large load a materially earlier connection date in exchange for a contractual obligation to reduce demand during a defined, capped number of scarcity hours. The obligation is enforceable and metered, with penalties for non-performance, not a voluntary best-efforts arrangement.
Grid capacity is sized for peak, and peaks are rare. A load willing to step back during the tightest hours can be accommodated on existing infrastructure that would otherwise require reinforcement. Modelling of the 22 largest US balancing authorities finds 76 to 126 GW of new load could be hosted if curtailed roughly 0.25 to 1% of hours — about 177 hours a year at 0.5% — led by PJM at 18 GW, MISO 15 GW, ERCOT 10 GW, and SPP 10 GW (Utility Dive summary of the Duke study).
Curtailable load avoids the capacity procurement that drives ratepayer cost shifts. It also fits AI training workloads, which are the most deferrable large computing task, unlike inference serving live traffic.
Google has signed roughly 1 GW of data centre demand response across five utilities as of March 2026 (Google).
Links
3Sub-issues
0Case studies
0