Authors: Adeyinka Olaseinde, Jonathan Frame – The University of Alabama
Title: Hydrologic Dependence of AI Data Centers Across U.S. River Basins
Presentation Type: Poster Presentation
Abstract: The rapid expansion of artificial intelligence (AI) data centers is emerging as a new anthropogenic influence on basin-scale hydrologic systems by concentrating industrial water demand within specific river basins. Despite their growing resource footprint, these facilities are not explicitly represented as distinct anthropogenic forcing in basin-scale water budgets. Here, we hypothesize that they should be. As large-domain models advance toward improved representation of human-water interactions, we ask whether AI infrastructure siting exhibits distinct hydrographic geospatial patterns within the context of basin-scale hydrologic characteristics. This could lead to better water forecasting, as we develop these anthropogenic influences as components in continental-scale hydrologic modeling frameworks such as the Next Generation Water Resources Modeling Framework (NextGen). We evaluate the hydrographic geospatial patterns of AI data centers across the contiguous United States using a national geospatial inventory and basin-matched stochastic baselines. Facility proximity to major river networks (HydroRIVERS ORD_STRA >= 5), lakes, and coastlines is quantified, and watershed stress is characterized using Aqueduct water-risk metrics. Chi-square tests of independence indicate that AI facilities are significantly overrepresented in high-stress basins relative to a baseline of N-matched random locations, with odds ratios approaching three for upper stress categories. In addition, basin-scale hydrologic regime metrics derived from National Water Model streamflow indicate that AI-hosting basins exhibit significantly lower flow variability and weaker seasonal concentration compared to the national baseline. While AI infrastructure shares certain siting tendencies with Toxic Release Inventory (TRI) sites, particularly in stress exposure, its association with hydrologically stable regimes suggests a distinct and measurable hydrographic structure relevant for next-generation hydrologic modeling.