AI Data Centers Increase Water Stress in Western US

Quartz reports that new AI-focused data center construction across the American West could require roughly 7 billion gallons of water per year, concentrating demand in Arizona, Nevada, and Utah. Quartz also reports Phoenix-area cooling withdrawals are projected to climb from about 385 million to more than 3.7 billion gallons annually as facilities proliferate. The Verge reports the proposed Stratos Project in Box Elder County, Utah, backed publicly by investor Kevin O'Leary, covers 40,000 acres and is reported to be projected to require up to 9 GW of power with early-phase costs above $4 billion. A Deseret News-Morning Consult poll shows 53% of Utah voters oppose the Box Elder proposal, and the Salt Lake Tribune reports Governor Spencer Cox criticized the rollout as "not good". Quartz cites company-level examples: Microsoft and Google operations each withdrew on the order of billions of gallons in recent reporting years, illustrating scale.

Data-center cooling choices drive the water problem. Industry reporting and technical analyses cited by Quartz show that much of the water withdrawn for evaporative cooling is consumed rather than returned to local systems, with 70-80% lost to evaporation in typical cooling-tower operations. Reporting across regional outlets notes alternative approaches, closed-loop systems, air-cooled designs, and geothermal cooling, are discussed by developers and local officials as lower-water options, though tradeoffs include higher energy use, cost, or reduced cooling efficiency in hot climates.

Industry and regional reporting frames the current situation as a clash among competing water users: municipalities, agriculture, tribal nations, and hyperscale infrastructure. Local political pushback is visible; Deseret News documents voter opposition in Utah, and multiple regional outlets report mounting community resistance in Wyoming and parts of Arizona and Nevada. Regulatory scrutiny has increased: state and county officials are facing questions about permitting timelines, water-rights transfers, and environmental review processes as developers seek large parcels with cheap power and available water rights.

For infrastructure planners and ML operations teams, the pattern matters because water availability is now a site-selection variable on par with power price and network latency. Reported numbers in Quartz and regional reporting indicate that cooling-driven water consumption can be material at hyperscale. Teams evaluating colocation or new build sites will increasingly need to factor water accounting, local regulatory risk, and community opposition into total-cost-of-ownership and resilience models.

• •Permitting and legal challenges in states with ongoing drought declarations, where county or state-level approvals may be delayed or conditioned.
• •Technology adoption: whether large projects commit publicly to low-water cooling technologies and how those choices affect energy consumption and build costs.
• •Tracking of disclosed water withdrawals in annual sustainability reports from major cloud and hyperscale operators, which Quartz and other outlets use as comparators.
• •Local political signals such as polling, county commission votes, and gubernatorial statements that can translate into moratoria or new rules.

Reporting from Quartz, The Verge, and regional outlets shows the rapid expansion of AI-oriented data centers is colliding with finite western water supplies. Editorial analysis: this is emerging as a practical constraint on siting and permitting for AI infrastructure, and practitioners responsible for capacity planning and site economics should treat water risk as quantifiable and trackable alongside power and connectivity metrics.