Regulated Large-Load Rules Create Financial Commitments without Mandating Equivalent Electricity Consumption

Abstract

Flexible computing loads (FCLs), including Bitcoin mining and some AI-training workloads, could absorb surplus generation and curtail during scarcity. Regulated prices and rate-of-return incentives may instead delay this shock-absorber response or convert FCLs into capital commitments. Whether these frictions eliminate operational flexibility remains empirically unresolved. This study manually audited official EIA-861 demand-response and dynamic-pricing records for 2013, 2020, and 2024 and final tariffs and orders for representative utilities in Washington, Georgia, Florida, Virginia, and Indiana. Commercial-and-industrial response served as an FCL proxy; financial obligations were separated from physical curtailment; utility fixed effects tested whether dynamic-pricing enrollment predicted realized-to-potential peak savings. In the only territory with a post-tariff EIA year, Puget Sound Energy reported 12.5 MW of actual 2024 peak savings from 13.3 MW potential (94.0%), after zero in 2013 and 2020; this descriptive change ranked at the 97.4th percentile among 39 Washington utilities. Across 105 utility-state-year observations, dynamic-pricing enrollment did not predict conversion (beta = -1.53 percentage points per enrollment point, SE = 3.22, p = 0.64). Georgia certified 9,985 MW of new generation, approximately 80% expected for data centers, while Florida, Virginia, and Indiana imposed 70%, 85%/60%, and 80% minimum-payment obligations. These rules allocate stranded-cost risk without mandating equivalent electricity consumption: Florida law expressly preserves reliability curtailment, and no reviewed order included customer computing hardware in utility rate base. Current evidence therefore rejects categorical prevention but does not identify an Averch-Johnson effect; regulation makes large loads capital-planning objects while shifting flexibility from automatic price response to contract-defined activation.