AI Data-Centre Water Evidence Explorer
Compare water withdrawal, estimated consumption, Water Consumption Impact and modelled off-site water across the tracker's ten-site seed. Every planning figure and unknown source remains visible as such.
Interactive front end over the frozen v0.5 dataset, published 8 August 2026. Water-source overlay checked 11 August 2026. DOI 10.5281/zenodo.21318960. CC BY 4.0.
Water Consumption Impact
Dimensionless peak consumptive draw as a share of host utility capacity.
One flat total can hide a change in source
Douglas County, Georgia is the current seed's explicit potable and reclaimed split. Total withdrawal changed little between the two reporting years, while the potable component increased and the reclaimed component declined.
This composition belongs beside WCI rather than inside it. The index measures total water burden against utility capacity; it does not grade the source.
Evidence table
| Site | Status | Water source | Cooling | Hydropower coupling | WCI |
|---|
Method and guardrails
WCI. WCI = (average withdrawal / host utility capacity) × consumptive ratio × peaking factor. A value is a pressure index, not a volume and not a water-quality score.
Consumption. Average consumption is withdrawal multiplied by the consumptive ratio. Ratios for non-Google rows are assigned framework defaults rather than site measurements.
Off-site share. This is a grid-average modelled band calculated from cooling WUE and the regional electricity-water intensity band. It is not a metered per-site value.
Source labels. Unknown means the current ledger does not support a classification. A reservoir in the hydropower field describes an electricity coupling and is not evidence of the site's cooling-water source.
Denominator. The seed remains ten rows: six primary-verified rows, three planning rows and one framework-secondary row. Filters change the display, never the denominator.