Thirsty servers are moving in on some of the nation's driest zip codes. Quartz reports that the AI-driven rush to build massive new data centers across the American West is sharply increasing water demand in regions already struggling with drought. In the Phoenix area alone, water use from cooling data centers is projected to jump nearly ninefold, from 385 million gallons annually to more than 3.7 billion gallons, much of it lost to evaporation through cooling towers. The surge comes as the Colorado River basin faces mounting strain, with Lake Mead about one-third full, Lake Powell below one-quarter capacity, and Arizona already absorbing mandatory cuts to its river allocation. The result is a growing collision between the tech industry's AI ambitions and one of the country's most stressed water systems.
Medium-size facilities can consume as much water annually as roughly 1,000 households, while hyperscale centers use far more. A regional study projected that data centers across Arizona, Colorado, Nevada, New Mexico, and Utah could consume 7 billion gallons of water annually by 2035, excluding the additional water needed to generate their electricity. The concerns extend beyond water. According to the Sierra Nevada Ally, Nevada data centers already account for roughly 22% of the state's electricity demand, with projections suggesting that share could grow sharply by 2030 as AI infrastructure expands. The report notes that a handful of planned projects could eventually require nearly three times the generating capacity of the Hoover Dam.
Elsewhere in the West, Quartz reports that Phoenix-area suburbs have adopted industrial water caps, Marana, Arizona, has barred data centers from using potable water entirely, and Tucson rejected a proposed project until developers pledged "zero-water" cooling technology instead. Pyramid Lake Paiute Tribe chairman Steven Wadsworth warned that developers are thinking too narrowly about long-term consequences: "Everybody's only focused on what's going to happen this year, and the next two, three years. And that's just not far enough to think."