Beyond the AI energy hype? data-center-related establishment growth, electricity demand, and emissions across U.S. states

Abstract

The rapid expansion of artificial intelligence (AI) and data-center infrastructure has raised concerns about growing pressure on electricity systems and the environmental consequences of rising computational demand. This study examines whether the expansion of data-center-related activity is associated with measurable changes in aggregate electricity demand and power-sector emissions across 50 U.S. states from 2010 to 2023. Using a balanced state-level panel and two-way fixed-effects models, data-center expansion is proxied by annual establishment counts, with population-normalized exposure and several alternative specifications used for robustness. The baseline estimates provide no evidence of a positive within-state association between growth in data-center-related establishments and aggregate electricity demand. This finding remains robust to alternative functional forms, population weighting, state-specific trends, lagged exposure, controls for grid investment and population, exclusion of major data-center states, winsorization, and first-difference estimation. Data-center activity is likewise not significantly associated with either total power-sector CO 2 emissions or average generation emissions intensity. These results should not be interpreted as evidence that data centers have no energy or environmental consequences. Rather, they highlight the limitations of annual state-level data and establishment-based measures in detecting infrastructure effects that may be highly localized and heterogeneous across facilities. The findings therefore underscore the importance of distinguishing aggregate electricity-system outcomes from localized grid constraints and average generation emissions from the marginal emissions associated with incremental data-center load.