Bilevel Analysis of Cost and Emissions Externalities from Data Center Load Shifting (2510.20805v1)

Abstract: Data centers are emerging as large, flexible electricity consumers capable of shifting computational workloads across locations in response to economic and environmental signals. While this flexibility has potential for emissions reduction, its impact on power system operations depends critically on how such behavior interacts with network constraints and market signals. We develop a bilevel optimization framework in which a data center minimizes a weighted combination of electricity cost and marginal emissions intensity (LME), while the system operator clears economic dispatch under transmission and generation constraints. Focusing on a stylized three-bus power system, we derive closed-form, piecewise-linear expressions for both the data center and system-wide objectives as functions of the data centers' load shift. These expressions capture threshold-driven regime changes due to congestion and renewable saturation. We identify sufficient conditions under which the data center's decentralized decisions align with or diverge from socially optimal behavior and characterize the resulting externalities. Our results reveal how system topology and generator asymmetry affect incentive alignment and provide insight into when marginal price or emissions signals may fail to guide flexible loads toward socially beneficial outcomes. Our results offer a tractable starting point for analyzing decentralized flexibility under carbon-aware incentives and suggest directions for improving coordination between flexible loads and system operations.