Evaluating the Economic and Operational Flexibility of Bitcoin Mining and AI Data Centers in the ERCOT Electricity Market

Abstract

The flexibility of large computing loads, the ability to ramp consumption up or down almost instantly without reducing operational efficiency, is empirically shown to respond quickly to grid conditions and electricity prices. Using an hourly ERCOT panel, a three-event study of grid stress, miner financial disclosures, and an engineering taxonomy of switching costs, we analyzed how flexible loads respond to market signals. Residual demand strongly predicts ERCOT electricity prices and scarcity across price thresholds from $80 to $200 per megawatt-hour. Riot Platforms' curtailment revenue, growing from $6.5 million in 2021 to $71.2 million in 2023, demonstrates how Bitcoin mining can profitably respond to these conditions. With AI data centers approaching 10 gigawatts of approved ERCOT capacity, the flexibility gap appears to be driven more by economic and workload characteristics than technical limitations. Mining hardware can be taken down virtually instantaneously at almost no cost, while AI inference and contractually bound cloud workloads are typically much less flexible because they are grounded to service-level agreements rather than the hardware itself. A hypothetical 100-megawatt Bitcoin mining dispatch counterfactual operating near the calculated break-even electricity price of $122 per megawatt-hour results in approximately 100,500 megawatt-hours of modeled curtailment and ~25,000–70,000 tons of potential avoided carbon emissions. Overall, the results suggest that workload characteristics and interruption costs, rather than hardware switching speed alone, determine the practical flexibility of large computing loads in ERCOT.