Voters made clear this week that they want limits placed on data centers for artificial intelligence, passing 40 of 41 ballot measures that would place limits on some of the more than 2,000 new facilities planned for communities across the country.
Water and power are top of ratepayers’ concerns. But calculating how much each data center is drawing from local resources is an elusive task, and operators typically report enterprise-wide usage figures rather than site-by-site.
Huge 1 GW AI data center such as meta, google, Amazon and SpaceX It could consume up to 10 million gallons of water a day, said Xiaolei Ren, a data center researcher and professor at the University of California, Riverside.
The exact numbers vary greatly depending on what kind of hardware is in the data center, where it is built, and what methods are used to cool the servers and the entire building.
The complexity of how water is used in data centers leaves room for misinformation from both builders and those who oppose it. CNBC spoke to experts to get a clearer picture.
Air cooling and liquid cooling
Data centers require serious internal cooling methods to prevent chips and servers from overheating.
Traditional data centers, filled with low-power servers used for things like networking and storage, are often air-cooled. Large fans move hot air away from the server and the building.
But modern AI data centers are equipped with high-power, high-density graphics processing units (GPUs) that can get hot enough to “boil an egg” on the servers, Ren said.
Water is about 3,000 times more efficient at cooling these GPUs, according to Christopher Wellies, Equinix’s head of sustainability.
GPUs “run at very high wattages and high temperatures, so if you were just using air, you would need hurricane force winds to really cool the GPU,” Wellise said in an interview with CNBC.
That’s why the latest rack-scale AI systems Nvidiawith vera rubin advanced micro device“Rely on liquid cooling, Helios. Most large AI data centers routinely use closed-loop liquid cooling systems that use very little water.
in Equinix CNBC got an up-close look at a closed-loop liquid cooling system being used to lower the temperature of an eight-row Nvidia Grace Blackwell rack owned by AI marketing startup Alembic at a data center in San Jose, California.
Equinix is building colocation data centers, where companies like Alembic rent space and operate their own server racks, and Equinix provides the power, network connections, interconnects, and cooling.
Like many data center operators, Equinix does not disclose water consumption for individual sites. In 2025, Equinix’s operations consumed 1.2 billion gallons of water, an increase of 10% from the previous year.
Despite closed-loop internal cooling like the one Equinix uses in San Jose, total water consumption is highly dependent on additional cooling processes that occur outside the data center.
“How do you shut off the heat from the closed loop to the surrounding environment?” Len said. “That’s where water usage becomes an issue.”
Meta, for example, is building a 1GW data center in Lebanon, Indiana that the company says will use “a closed-loop liquid cooling system that recirculates the same water” and will use no water for most of the year.
But the company’s law firm says water allocations for the site could increase to 8 million gallons per day by 2031. Mehta said the numbers represent peak usage on the hottest days. “We expect this maximum usage level to be reached rarely, and is currently projected to occur only a few days per year,” Mehta told CNBC in a statement.
That’s because on the hottest days of the year, the external cooling process requires more water and removes heat from another internal closed-loop system.
More power with less water
There are two main methods for this external cooling process in data centers: dry cooling and evaporative cooling. Dry cooling uses much less water, but can require much more power, especially in the hot climates where many AI data centers are built.
“There’s no such thing as a silver bullet,” Wellies said. “At some point, you have to get that heat out into the atmosphere, so there’s a trade-off.”
Dry cooling uses a fan to move air across the heat exchanger and uses the surrounding air to cool the heat exchanger. High outside temperatures make these fans less effective, so the data center first turns on the cooling equipment to cool the air.
“This is a very power-intensive process, and that’s exactly why we run into power shortages in the summer,” Ren said. “Because everyone else has air conditioning on.”
Evaporative cooling, on the other hand, consumes very little electricity. The hot water circulates over a cooling tower or a moist pad exposed to the air, and the evaporating portion also carries away the heat. On the hottest days, large amounts of water can be lost through evaporation.
Hybrid approaches are common as companies seek to strike a balance between these two extremes.
Some data centers, like Equinix in San Jose, use dry cooling during cold weather and evaporative cooling when it’s hot outside. In Indiana, Meta uses dry cooling, which uses mist to cool the air during hot weather rather than relying solely on refrigeration equipment to reduce power consumption.
Creating cooler chips
Near some data center projects, Meta, Amazon, Microsoft and others are building water treatment plants to treat and recycle the water they need. For example, Meta spent $100 million on a new water and wastewater system near its data center in Kuna, Idaho.
Elon Musk has pledged $80 million to build a water recycling plant outside Memphis, Tennessee, that will turn treated municipal wastewater into cooling water for SpaceX’s Colossus data center. The first data center there, Colossus 1, quickly came online in 2024, but construction of the water treatment plant was postponed to April of this year.
Colossus 1 is allocated up to 1.28 million gallons of water per day, but has not yet reached that threshold, according to local utility company Memphis Light Gas & Water.
Tech companies are also building their own power plants at some data center campuses to ensure adequate supplies, but they also consume large amounts of water. Colossus, which derives its energy from gas turbines installed on-site, is the subject of multiple lawsuits for alleged violations of the Clean Air Act and noise pollution.
Meanwhile, some chip companies are working directly on ways to make chips cooler so that they require less cooling in the first place.
South Korean startup DeepX makes AI processors that run up to 100°F cooler than competitors. hyundai car DeepX chips are used in robotic products.
SK HynixThe world’s leading manufacturer of high-bandwidth memory told CNBC in July that it was working on a low-heat memory technology called iHBM that could reduce thermal resistance by 30%.
At Equinix, Wellies said the company conducts water risk analyzes at each new site to determine which external cooling methods to install in areas where water is scarce.
“Communities are starting to ask the right questions. They’re concerned about their water consumption,” Wellies said. “We take resource consumption, resource utilization very seriously.”
WATCH: Race to power AI data centers is reshaping America’s power grid
