Answers are grounded in this paper and cite their sources.
Introduction Growth in cloud services and artificial-intelligence computing has intensified policy attention to the physical infrastructure of data centres. The empirical exposure is not AI-specific: PeeringDB does not consistently identify workload, GPU or HPC deployment, IT load, electricity capacity, cooling demand, or the share of a facility devoted to AI-related computation. The study therefore examines ecological associations with current PeeringDB-listed facility records while retaining a broader set of panel, spatial, distributional, and facility-centred analyses to assess uncertainty and heterogeneity. Methods The study uses 200 fixed-2025 GHSL Urban Centre polygons and 1,400 annual observations from 2019 to 2025. Facility count means the number of current PeeringDB-listed facility records with valid coordinates inside the relevant polygon. The primary models relate log(1 + facility count) to 2025 NDVI, McFeeters NDWI, and NDBI, condition 2025 outcomes on their corresponding 2019 values, and examine 2023–2025 means. Two-part models separate matched-record presence from the listing-count gradient among 131 positive-listing centres. Secondary analyses retain the repeated panel, Urban Centre fixed-effects trajectories, alternative exposure and sample definitions, bootstrap and multiple-testing checks, regional, quantile, hierarchical, and spatial models, and 2 km and 5 km profiles around 26 audited facilities. Seven facilities have directly verified timing for descriptive event profiles. Results The primary full-sample models do not show significant associations between current facility count and 2025 NDVI, McFeeters NDWI, or NDBI, and the conclusion is unchanged after the 2019 starting-condition adjustment and in 2023–2025 means. Matched-record presence is also null. Among positive-listing centres, the simple 2025 count gradient is positive for NDVI and negative for NDWI, but the NDWI coefficient attenuates from −0.021 ( p = 0.006) to −0.002 ( p = 0.243) after 2019 NDWI is included. In the repeated 2019–2025 panel, the global NDWI coefficient remains borderline and the fixed-effects trajectory models do not show a stable adverse pattern. Distributional, regional, and spatial models reproduce a negative conditional NDWI association in some specifications, while the seven directly timed facilities show no common post-opening satellite signature. Discussion The primary design does not detect a robust adverse ecological association between current PeeringDB facility count and near-contemporaneous city-wide surface conditions. The broader analytical package is retained because it shows where patterns recur, where they disappear, and which conclusions depend on modelling choices. The conditional NDWI association remains mechanistically unresolved and is substantially connected to earlier NDWI conditions. None of the analyses identifies facility openings, AI-specific effects, cooling-water withdrawal, groundwater depletion, or municipal water stress. City-wide non-detection also cannot establish safety or exclude parcel- and neighbourhood-scale effects.
No takes yet. Share an insight, caveat, or question.
NICOLE DSILVA (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: