AI-driven data centers are drawing growing opposition from communities across the United States as residents raise alarms over water withdrawals, energy consumption and pollution. Protests have appeared from Texas to New Mexico and Arizona, where locals say new server campuses could deepen stress on limited supplies. The criticism focuses on evaporative cooling systems and power generation tied to fossil fuels, both of which can increase local water demands.
Cooling modern processors is energy- and sometimes water-intensive: server components can run very hot and some facilities still use evaporative towers that lose water to the atmosphere. Estimates vary widely. Researchers have calculated that data center cooling used about 66 billion liters of water in 2023, under 1 percent of national consumption, yet forecasts by Fengqi You and colleagues project between 731 billion and 1,125 billion liters annually in the United States by 2030 under high-demand scenarios. Earlier per-query water figures—such as a 2024 estimate of roughly 500 ml for a short AI response—have been revised downward as models and systems become more efficient; for example, one large cloud provider later estimated only a few drops per typical query.
Engineers and researchers point to several levers that could sharply reduce AI’s water footprint. Shifting electricity supply from steam-cooled fossil generation to wind and solar cuts the indirect water used to cool turbines. Strategic siting of facilities in water-abundant regions — names often cited include Montana, Nebraska, parts of Texas and South Dakota — can avoid adding stress in drought-prone states such as Arizona, New Mexico and California. Technological shifts also matter: many firms now deploy liquid cooling and are testing immersion systems to limit evaporative losses. Major providers such as Amazon, Microsoft and Google are rolling out such approaches, while university researchers and companies explore immersion and other advanced options.
Corporations have announced water stewardship targets: OpenAI describes plans to limit onsite water use, Google aims to replenish 120 percent of freshwater its centers use, and Amazon reports progress toward being water positive by 2030. Reusing waste heat for district heating in Europe offers another way to boost sustainability. As Eric Masanet and other experts note, the trajectory of AI’s water demand will depend on choices about siting, energy sources and cooling design — decisions that can either magnify local stress or dramatically curb future consumption.
