Title:
ARTIFICIAL INTELLIGENCE, DATA-CENTRE EXPANSION, AND UNITED STATES GRID RELIABILITY: NONLINEAR CAPACITY CONSTRAINTS, INFRASTRUCTURE RESILIENCE, AND CROSS-STATE SPILLOVERS |
Abstract:
The rapid expansion of artificial intelligence (AI) and data-centre infrastructure is reshaping the electricity-demand outlook of the United States and creating new challenges for power-system reliability. This study examines whether the reliability effects of AI-driven electricity demand vary according to grid-capacity conditions and whether infrastructure resilience can moderate these effects. Using a simulated panel of 500 state-year observations representing the 50 U.S. states, the study applies fixed-effects estimation, a Panel Smooth Transition Regression (PSTR) model, a Spatial Durbin Model (SDM), and Dynamic Common Correlated Effects (DCCE) estimation. The results indicate a positive and statistically significant association between AI-driven electricity demand and grid-reliability stress. Importantly, the PSTR estimates reveal substantial nonlinearities: the estimated coefficient increases from 0.176 in the lower-pressure regime to 0.612 in the higher-pressure regime, with a transition point of approximately 15.8% reserve margin. Renewable-energy penetration, battery storage, and grid investment exhibit significant buffering effects, with battery storage showing the strongest simulated moderating effect. Spatial estimates further indicate that AI-related electricity pressure extends beyond individual states, demonstrating the importance of interconnected regional electricity systems. The findings suggest that AI expansion should be treated as a capacity, resilience, and regional-coordination challenge, rather than solely as an issue of aggregate electricity consumption. For U.S. federal and state authorities, the study supports capacity-sensitive data-centre siting, accelerated transmission and storage investment, large-load flexibility requirements, improved AI-energy data reporting, and stronger coordination among DOE, NERC, FERC, RTOs/ISOs, utilities, and state regulators. Because the analysis uses simulated observations, the numerical estimates are illustrative and require validation using observed state-level data.
|
