data-center cooling and its electricity supply draw freshwater from basins that are already stressed

Two distinctions that decide the argument

Withdrawal is not consumption. Withdrawal is water taken from a source; consumption is withdrawal minus what is returned. US off-site electricity-embedded water averages 43.8 L/kWh withdrawn but only 3.1 L/kWh consumed. On-site cooling ranges 1 to 9 L/kWh, from about 1 at Google's global average to 9 at a large Arizona facility in summer (Li, Yang, Islam and Ren, arXiv 2304.03271). Conflating the two produces figures off by an order of magnitude in either direction.

Indirect water exceeds direct water. Lawrence Berkeley National Laboratory estimated US direct data-center cooling water at 17 billion gallons in 2023 against roughly 200 billion gallons embedded in the electricity consumed (Latitude Media). Modelling for Phoenix projects electricity-related water consumption rising ~400% (2.9 to 14.5 billion gallons/year) alongside cooling water rising ~870% (Ceres, Drained by Data). A facility that eliminates on-site water has not eliminated its water footprint — it has moved it to the generating station and may have increased it.

Aggregate figures mislead in both directions

Arizona's entire data-center sector used about 905 million gallons in 2025 — under 0.1% of state consumption — and CyrusOne's eight Chandler facilities use no cooling water, only ~180,000 gallons/year for humidification (Circle of Blue). At the same time, Google's campuses in The Dalles, Oregon reached ~550 million gallons/year, close to 40% of that city's total consumption (Latitude Media).

Both facts are true. The harm is basin-scale and utility-scale, never national. Across 632 Amazon, Microsoft, and Google facilities, 38 active sites plus 24 in development sit in water-scarce areas; Google's first Mesa, Arizona centre is permitted for 5.5 million cubic metres/year — equivalent to ~23,000 residents (SourceMaterial / Guardian).

Who is affected

Users on the same municipal system or aquifer, and agriculture competing for the same allocation. The hot, dry climates where evaporative cooling performs best are frequently the most water-stressed.

Where causation is contested

In Morgan County, Georgia, residents whose wells sit ~1,000 feet from Meta's campus report sediment and pressure loss; one household has spent ~$5,000 on appliances with ~$25,000 needed for a new well. The facility uses ~500,000 gallons/day and the county projects a water deficit by 2030 (NYT via reprint). Meta commissioned a groundwater study concluding the facility was "unlikely" to be responsible; no independent hydrogeological assessment has been published (CBS Atlanta).

Scope

This issue covers freshwater withdrawal and consumption attributable to data-centers — direct and indirect — in the basins where it occurs. Corporate water accounting and site-level disclosure are handled under a separate sub-issue, because replenishment credits in one watershed do not relieve scarcity in another.

Impact if unaddressed

Permits are being issued at volumes equivalent to small towns, in basins where the counterfactual use is drinking water and irrigation, on the basis of aggregate statistics that cannot detect the harm. Where design was challenged early it changed: two large projects in South America were reformulated to air cooling. Where it is not challenged, the design is fixed for the 20–30 year life of the facility.