Hidden Currents: A Peak-Stress Water Index for Auditing AI Data Center Sustainability and Informing Disclosure Policy

Data center water use is usually reported as Water Usage Effectiveness (WUE), an annual, fleet-averaged ratio of water to IT energy. WUE hides two dimensions of impact on host communities: the concentration of demand on peak summer days and the scarcity of water where a facility is located. Prior work has examined these dimensions separately. We propose the Peak-Stress Water Index (PSWI), the product of annual WUE, a peaking factor β (peak-day to average-day water use), and the facility’s Baseline Water Stress (BWS) from WRI Aqueduct 4.0. We apply PSWI to an open dataset of 48 hyperscale and colocation facilities run by seven operators in North America, Europe, and Asia, using only WUE values that operators publish or that can be derived from their published totals. Because none of these operators publishes facility-level WUE, facilities inherit fleet or regional values, so PSWI isolates how location changes the burden behind the same headline number. PSWI spans a factor of about 1,500 across facilities; within one operator that applies a single fleet WUE, it differs 267-fold between Henderson, Nevada and Hamina, Finland. Company rankings by mean PSWI depart from rankings by operator WUE (Spearman ρ = 0.71), although bootstrap intervals for individual company shifts include zero. Our climate-based β is strongly correlated with BWS (r = 0.84); removing it leaves facility and company rankings nearly unchanged (ρ = 0.96 for both), so the results are driven mainly by location stress. In an idealized reallocation exercise, moving 10% of workload from the highest-to the lowest-PSWI facilities lowers aggregate PSWI by 35% (95% interval 21–46% under assumed input uncertainty), showing that impact is concentrated in a small tail. None of the audited operators publishes peak or monthly water data. We outline a facility-level disclosure framework that borrows the phase-in architecture of U.S. securities disclosure rules but could be implemented through state and local water permitting. Data and code are openly released.