This week, voters clearly expressed their desire to impose restrictions on artificial intelligence data centers. Of the 41 proposed voting measures, 40 were passed, and these measures will impose restrictions on some of the more than 2,000 new facilities planned to be built in communities across the country.

What residents are most concerned about is water and electricity. However, calculating exactly how much each data center consumes from local resources is actually a very difficult task, as operators usually disclose their usage on a company-wide basis rather than per individual site.

According to Shaolei Ren, a researcher and professor at the data center of the University of California, Riverside, large data centers with a capacity of 1 gigawatt from companies such as Meta, Google, Amazon, and SpaceX can consume up to 10 million gallons of water per day.

The specific numerical differences are quite significant and depend on what type of hardware is installed within the data center, where it is located, and what methods are used to dissipate heat from the servers and the entire building.

The water usage methods in data centers are complex, which also provides an opportunity for both proponents and opponents to spread incorrect information. CNBC interviewed several experts to gain a clearer understanding of this issue.

Air cooling vs. liquid cooling

Data centers require powerful internal cooling systems to prevent chips and servers from overheating.

Traditional data centers are equipped with servers that consume less power and are typically used for networking or storage purposes, often utilizing air cooling systems. Large fans are used to carry away the hot air generated by the servers and expel it outside the building.

But modern AI data centers are filled with high-power-consuming, higher-density graphics processors, namely GPU. Ren says that these GPU are so hot that they can even "cook an egg".

Equinix The person in charge of sustainable development, Christopher Wellise, stated that water is about 3,000 times more efficient at cooling these GPU than air.

In an interview with CNBC, Wellise said, "The power and temperature of GPU are both very high. If you only use air to cool it down, you would almost need wind forces at hurricane level to truly bring them down."

This is also why the latest rack-level AI systems – NVIDIA’s Vera Rubin, Advanced Micro Devices, and Helios – rely on liquid cooling. Most large AI data centers use closed-loop liquid cooling systems, which consume very little water on a daily basis.

At a Equinix data center in San Jose, California, CNBC took a close look at this closed-loop liquid cooling system. This system is used to cool a row of 8 Nvidia Grace Blackwell racks owned by Alembic, a marketing startup founded by AI.

Equinix builds managed data centers, where companies like Alembic can rent space to operate their own server racks, while Equinix provides power, network connectivity, interconnection, and cooling services.

Like many data center operators, Equinix does not disclose the water consumption of individual sites. In 2025, Equinix consumed a total of 1.2 billion gallons of water in its operations, an increase of 10% from the previous year.

Although an internal closed-loop cooling system like the one used at Equinix in San Jose can reduce water consumption, the total amount of water used depends to a greater extent on the additional cooling processes outside of the data center.

Ren said, "How do we release the heat from a closed-loop system into the environment? The use of water is where this comes into play."

For example, in Lebanon and Meta of Indiana, a 1-gigawatt data center is under construction. The company stated that the center uses a "closed-loop liquid cooling system" that recycles the same batch of water and does not use water for most of the year.

However, according to the company's law firm, the water quota for this project may increase to 8 million gallons per day by 2031. Meta indicates that this figure represents the peak water consumption under the hottest weather conditions. "We expect to rarely reach this highest level of water consumption; currently, it is estimated that this will only occur a few days throughout the year," Meta told CNBC in a statement.

The reason is that the hottest days of the year require more water for the external cooling process to carry away the heat from the internal closed-loop system.

Use less water, use more electricity.

There are mainly two methods of external cooling for data centers: dry cooling and evaporative cooling. Dry cooling uses less water, but it may require more electricity, especially in hot regions where many AI data centers are being constructed.

Wellise said, "There is no magic cure. In the end, you will need to release that heat into the atmosphere, and that is where the trade-offs occur."

Dry cooling relies on fans to blow air over heat exchangers, using ambient air to cool down the system. However, when the outdoor temperature is very high, the effectiveness of these fans decreases, so data centers will activate cooling units first to lower the air temperature.

Ren said, "This is a very power-consuming process, and that's exactly why we experience power shortages in the summer, because everyone else is also using air conditioners."

In contrast, evaporative cooling consumes very little electricity. Hot water circulates within the cooling tower or flows over wet pads exposed to the air, and the part that evaporates takes away the heat. However, on the hottest days, a large amount of water is lost through evaporation.

Between these two extremes, hybrid solutions are quite common, as companies attempt to find a balance between them.

Some data centers, such as the facility of Equinix in San Jose, use dry cooling in colder months and switch to evaporative cooling during hot weather. In Indiana, Meta relies on dry cooling and uses spraying to cool the air in hotter months, rather than relying solely on refrigeration units, in order to reduce power consumption.

Make chips cooler

Near some data center projects, companies such as Meta, Amazon, and Microsoft are building water treatment plants to process and recycle the required water. For example, Meta spent $100 million near Kuna in Idaho for a new water and wastewater system.

On the outskirts of Memphis, Tennessee, Elon Musk promised to invest $80 million in building a water recycling plant that is intended to convert treated municipal wastewater into cooling water for a data center with the code name SpaceX Colossus. The local first data center, Colossus, was quickly put into operation in 2024, but the construction of the water treatment plant was postponed in April of this year.

According to the local utility companies Memphis Light, Gas, and, and Water, the maximum water quota for Colossus 1 can reach up to 1.28 million gallons per day, but this limit has not yet been reached.

Tech companies are also building their own power plants within some data center parks to ensure they have sufficient electricity, but these facilities also consume a large amount of water. Colossus uses on-site installed gas turbines for power generation, and these devices have faced multiple lawsuits for alleged violations of the Clean Air Act as well as noise pollution.

At the same time, some chip companies are also directly researching how to reduce chip heating, thereby reducing the need for cooling from the source.

A AI processor produced by a South Korean startup DeepX is claimed to operate at a temperature up to 100 degrees Fahrenheit lower than that of its competitors. Hyundai Motor will use the DeepX chip in its robotic products.

Global leading high-bandwidth memory manufacturer SK Hynix told CNBC in July that the company is developing a low-heat-generation memory technology called iHBM, which can reduce thermal resistance by 30%.

In Equinix, Wellise indicates that for each new site, a water resource risk analysis is conducted first to determine which external cooling method to adopt in areas with scarce water resources.

Wellise said, 'The community has started to ask the right questions. They are concerned about water consumption. We take resource consumption and usage very seriously.'

WATCH : The race to power AI data centers is reshaping America 's electric grid