Behind-the-Meter Versus Grid Electricity for AI Datacenters: Magnitude, Geography, and Equity of PM_2.5-Attributable Mortality in Memphis

Background: A growing number of datacenters are powered by colocated, behind-the-meter (BTM) fossil-fuel generation rather than the public grid. The exposure to air pollution and the health impacts of BTM supply have not been directly compared to the conventional grid supply. Objective: Using Colossus 1 (C1) and Colossus 2 (C2), two datacenters 12 km apart in the Memphis region, we compared the magnitude, spatial distribution, and equity pattern of fine particulate matter (PM2.5)-attributable mortality associated with grid versus BTM supply. Methods: We evaluated two scenarios: (i) a scenario in which C1 and C2 each hypothetically used 240 MW; and (ii) a real-world scenario based on the reported capacity of C1 and the permitted capacity of C2. We estimated population-weighted PM2.5 exposure and attributable mortality by distance, household income, and strata of people-of-color. Results: Under the same 240-MW demand, C1 had higher PM2.5-attributable mortality than C2: 10.83 deaths per year (95% interval, 7.83-13.84) versus 3.67 (2.65–4.68). In the real-world scenario, the order reversed: 11.35 (6.48-17.29) for C2 versus 8.10 (4.60-12.38) for C1. Statistically, 94% of C1 mortality occurred at least 250 km from the facility, while 82% of C2 mortality occurred within 50 km of the facility. The low-income, high-people-of-color stratum accounted for the largest share of C2-attributable mortality. Conclusions: The electricity-supply pathway shaped the magnitude and spatial distribution of PM2.5-attributable mortality. The grid supply dispersed mortality across distant communities, whereas the BTM generation concentrated it near the facility. Our findings support integrating epidemiological evidence into datacenter siting and permitting so that potential public-health consequences can be identified and avoided before decisions are made.