Data Centers & Water

The Thirsty Cloud

AI and cloud computing are consuming millions of gallons a day in the same drought-stricken western states this site tracks.

PCSN is an independent citizen science project, not a government agency. Figures below are drawn from published public sources and stated conservatively.

Every AI query, cloud backup, and streaming video routes through a data center, and those buildings run hot. To keep servers from overheating, many use evaporative cooling, which consumes water directly. A single large hyperscale facility can use on the order of 1–5M gallons per day, comparable to a town of 10,000 to 50,000 people (industry and utility reporting; see Sources).

The problem is where this demand is landing. New data centers are increasingly sited in the arid West, Arizona, Nevada, Utah, and eastern Oregon and Washington, the exact drought corridor tracked on our Western Water Crisis page, where 39 reservoirs across 11 states sit near record lows. As of August 2026, Lake Powell is around 23% of capacity and Lake Mead around 28%, near a record low. Growing computing demand and shrinking water supply are converging on the same ground.

The Scale of the Draw

Data centers consume water two ways: directly, through evaporative cooling on site, and indirectly, through the water used to generate their electricity at thermoelectric power plants. Both draw on the same stressed western basins. The figures below are national estimates; local impacts depend on siting and cooling design.

~17B gal

Direct US water use, 2023

~66 billion liters (LBNL 2024)

1–5M gal/day

One hyperscale facility

A town of 10k–50k people

2 pathways

Direct + indirect draw

Cooling plus power generation

Why AI Is Accelerating This

Training and serving large AI and machine-learning models is unusually energy-intensive, and every watt of electricity that becomes heat has to be carried away. Where that heat is removed by evaporative cooling, it is paid for in water. Peer-reviewed work, including Li et al. (2023), Making AI Less Thirsty, and Ren et al., documents a rising water footprint as AI workloads scale. Because total water use also includes the water consumed generating the electricity, the full footprint is considerably larger than the direct on-site figure alone. We report these numbers as estimates because measurement methods and corporate disclosure still vary widely, see our Methods page for how we treat uncertainty.

Common Questions

How much water does a data center use? A large hyperscale data center can consume on the order of 1 to 5 million gallons of water per day for cooling, comparable to the daily use of a town of roughly 10,000 to 50,000 people. Exact figures vary widely with cooling design, climate, and reporting method; the Lawrence Berkeley National Laboratory 2024 US Data Center Energy Usage Report finds hyperscale facilities dominate national water use.

How much water do US data centers use overall? The LBNL 2024 report estimates US data centers consumed on the order of 66 billion liters (roughly 17 billion gallons) of water directly in 2023. Total water use is substantially higher once the indirect water used to generate their electricity is counted.

Why does AI increase data-center water demand? Training and running large AI and machine-learning models is energy-intensive, and the servers involved generate heat that must be removed, often through evaporative cooling that consumes water directly, plus additional water used at the power plants supplying electricity. Peer-reviewed work such as Li et al. (2023), Making AI Less Thirsty, documents this growing footprint.

Where are these data centers being built? Many new facilities are sited in arid western states, Arizona, Nevada, Utah, and eastern Oregon and Washington, that are already in drought. These are the same basins this site tracks, where reservoirs like Lake Powell and Lake Mead have fallen to a fraction of capacity.

Sources: Shehabi et al., 2024 United States Data Center Energy Usage Report, Lawrence Berkeley National Laboratory (2024); Li, Yang, Islam & Ren, Making AI Less Thirsty: Uncovering and Addressing the Secret Water Footprint of AI Models (2023); USGS water-use data. Reservoir figures reflect this site’s Western Water Crisis tracking (39 reservoirs, 11 states; spring 2026). Water-use figures are national estimates stated conservatively; local impacts depend on facility siting and cooling design. PCSN is an independent citizen science project, not a government agency.