Published: June 05, 2026, 01:13 UTC
SpaceX did something unusual in its IPO filing this spring. Alongside the standard risk factors about launch failures and government contracts, the company flagged water access as a constraint on data center development. It was a sign that the technology industry’s water problem has moved from an environmental concern to a business risk.
Data centers consume enormous amounts of water, primarily for cooling. A typical facility uses about 1.1 million liters (300,000 gallons) per day. A large AI-focused data center can use several times that. Google’s Council Bluffs, Iowa facility alone consumed over 3.8 billion liters (1 billion gallons) in 2024. Training GPT-4 required approximately 600 million liters (158.5 million gallons) of water (Ars Technica; WIRED).
A Gallup poll conducted earlier this year found that seven in 10 Americans oppose new data center development in their communities, with water consumption cited as the top concern (Gallup).
The scale of the problem
Global data center water consumption stands at approximately 560 billion liters (148 billion gallons) per year. Bloomberg and United Nations projections put that figure at 1.2 trillion liters (317 billion gallons) by 2030, more than double current levels (Bloomberg; UN).
In the United States alone, data centers consumed approximately 66 billion liters (17.4 billion gallons) directly in 2024, with indirect water use from electricity generation adding another 800 billion liters (211 billion gallons), according to Lawrence Berkeley National Laboratory (LBNL). Texas data centers used roughly 185 billion liters (49 billion gallons) in 2025, with no statewide tracking mechanism in place (HARC).
AI workloads are making the problem worse. Research from UC Riverside found that AI-focused data centers use 10 to 50 times more cooling water than traditional server farms. Each ChatGPT query consumes roughly 0.3 milliliters of water, according to OpenAI CEO Sam Altman, a figure that, while tiny per request, adds up across billions of daily interactions (IEEE Spectrum).
No single solution
The industry is pursuing multiple cooling strategies, each with trade-offs.
Evaporative cooling remains dominant, used in roughly 83% of data centers. It is energy-efficient, consuming less electricity than alternatives, but requires large volumes of fresh water. Average water usage effectiveness is about 1.9 liters per kilowatt-hour (EESI). UC Riverside researcher Shaolei Ren has pointed out that evaporative cooling can free up 10 to 30 gigawatts of grid power at peak times because it uses less electricity than dry cooling.
Direct-to-chip liquid cooling circulates fluid through cold plates attached directly to processors. It accounts for 47% of the liquid cooling market and is growing at 12.4% annually, according to Mordor Intelligence. Microsoft began fleet deployment in July 2025 (Mordor Intelligence).
Immersion cooling submerges servers in dielectric fluid. Companies including Submer and LiquidStack offer single-phase and two-phase systems, though adoption remains limited to specialized deployments.
Zero-water cooling uses closed-loop systems that recycle coolant without consuming fresh water. Microsoft says its zero-water design saves 125 million liters (33 million gallons) per facility per year and will become standard from late 2027. At Build 2026, the company claimed its newest AI data center uses water at levels comparable to a restaurant kitchen (TechRadar; Microsoft).
Dry cooling achieves water usage effectiveness of zero but requires significantly more electricity, which can increase carbon emissions depending on the local grid mix.
The water-energy trade-off
The fundamental tension is that saving water often means burning more electricity, and vice versa. Evaporative cooling uses less grid power because water absorbs heat more efficiently than air. Switching to dry cooling reduces water consumption but increases energy demand and, in many regions, carbon emissions.
Google’s approach reflects this complexity. “A one-size-fits-all strategy just does not work,” Ben Townsend, Google’s head of data center water strategy, told WIRED. The company uses site-specific designs: reclaimed water in water-stressed regions and evaporative cooling where water is abundant. Google has committed to replenishing 120% of its water consumption by 2030 (Axios).
Microsoft has pledged to be water positive by 2030, aiming for a 40% improvement in water intensity. But internal documents obtained by the New York Times show the company’s water use jumped 34% amid the AI buildout and is expected to double by 2030 from 2020 levels (NYT).
AWS is expanding its use of reclaimed water, currently deployed at 20 sites with a target of 120 by 2030. The OpenAI-Oracle Stargate project, a $500 billion joint venture, is moving away from evaporative cooling entirely, though the specific technology has not been disclosed.
Regulatory pressure builds
Several US states are moving to regulate data center water use. California’s AB 2469 would require water disclosure for large facilities. Virginia’s Data Center Reform Coalition is pushing for efficiency standards. Texas, which consumed roughly 185 billion liters (49 billion gallons) from data centers in 2025, has no tracking mechanism in place, leaving a significant gap in oversight.
Priscilla Johnson, a former Microsoft data center executive, told WIRED that external pressure has been necessary to drive change. “The industry has to be challenged to design smarter,” she said.
The water problem is not going away. Every new AI data center built today will consume water for a decade or more. The companies that solve this tension between compute growth and water constraints will have a structural advantage. Those that ignore it may find their next data center blocked by local opposition, regulatory action, or simple physics.
Sources: Ars Technica/WIRED (June 4); Bloomberg (2025); New York Times (January 27); Axios (June 3); IEEE Spectrum (2025); Lawrence Berkeley National Laboratory; EESI