How much water do data centers consume annually in the United States, and what percentage of local water supplies does this represent in high-growth regions?

Data centers power everything from social media and streaming to AI tools like chatbots, but they need huge amounts of water to stay cool. In 2021, U.S. data centers used about 449 million gallons per day—roughly 164 billion gallons a year. That’s only 0.14% of all water withdrawn nationwide, far less than farming or power plants.

Cooling systems often rely on evaporative methods, where water absorbs heat and then evaporates, which is efficient but removes water from local supplies. A medium-sized facility can use 110 million gallons annually, similar to what 1,000 households consume. Larger 100-megawatt centers may draw 2 million liters daily. While these numbers look small nationally, they concentrate in fast-growing areas like Newton County, Georgia, or Sun Belt states where facilities cluster near cheap power and internet cables. There, data centers can take 1–5% or more of municipal water during peak times, straining supplies especially as AI workloads increase demand.

Some experts and industry voices point out that many centers recycle water or switch to air cooling, and they bring jobs plus tax revenue that support local economies. Others, including environmental groups, argue that even modest shares can worsen shortages in drought-prone regions and that current rules don’t require enough transparency or efficiency upgrades.

You rely on these facilities daily, so the trade-offs affect both the apps you use and the water available in your community long-term. Policies like mandatory water reporting and incentives for recycled or non-evaporative cooling aim to balance growth with conservation.

Data center water consumption has emerged as a pressing policy concern amid the expansion of artificial intelligence infrastructure across the United States. Nationally, the CAES Field Report estimates that facilities withdrew approximately 449 million gallons per day in 2021, or roughly 164 billion gallons annually. This volume represents only 0.14 percent of total U.S. water withdrawals, a figure dwarfed by agricultural and thermoelectric uses according to Congressional Research Service analyses. Yet these aggregate statistics conceal pronounced regional pressures where hyperscale campuses cluster near affordable power and fiber routes. A medium-sized facility may consume 110 million gallons yearly—comparable to the demand of about one thousand households—while a 100-megawatt installation can require two million liters daily, with peaks reaching five million gallons under extreme conditions.

This disparity between modest national shares and concentrated local impacts raises difficult trade-offs. In water-stressed Sun Belt counties, proposed projects could claim 1–5 percent of municipal supplies during peak seasons, according to World Resources Institute assessments. Evaporative cooling technologies, which dominate existing stock, lock operators into high withdrawal rates even as AI workloads intensify thermal loads. Conversely, proponents emphasize economic returns—employment, tax revenue, and productivity gains from advanced computing—while noting that many operators already recycle water or adopt air-side economization. The Environmental and Energy Study Institute highlights successful shifts toward non-evaporative systems, yet capital costs and performance guarantees slow adoption.

Policy responses therefore center on mandatory water reporting, efficiency benchmarks, and incentives for closed-loop cooling and recycled supplies. Such measures aim to improve transparency without stifling innovation, though implementation faces hurdles: inconsistent metering standards, limited state planning capacity, and the difficulty of retrofitting legacy facilities. Theoretical considerations of distributive justice further complicate siting decisions, as benefits accrue nationally while scarcity risks fall on specific communities. Evidence from Lincoln Institute case studies in Georgia illustrates how infrastructure can accommodate current loads yet become strained under successive expansions. Without coordinated standards that integrate data centers into regional water budgets, high-growth clusters risk localized shortages even as aggregate national consumption remains modest.