The Truth About How Data Centers Use Water and Power to Keep Cool

How Data Centers Use Water and Power to Cool AI Systems

The rapid expansion of artificial intelligence is creating a growing demand for electricity and water across the United States. Behind every AI-generated image, chatbot response and complex computing task is a network of powerful computer chips operating inside massive data centers. Keeping those machines from overheating requires sophisticated cooling systems that can consume significant amounts of electricity and, depending on the technology used, substantial quantities of water.

As technology companies race to build larger facilities to support AI, questions about water consumption, electricity costs and pressure on local infrastructure are becoming increasingly important. Communities are questioning how much water these facilities actually use, whether they could affect household utility bills and what happens when massive data centers are built in areas already facing water shortages.

However, the relationship between data centers and water consumption is more complicated than it may initially appear. Many modern facilities use closed-loop liquid cooling systems that circulate coolant rather than continuously consuming fresh water. That does not mean the facilities operate without water. Their external cooling infrastructure can still require significant resources.

How Data Centers Actually Use Water to Cool AI Chips

Closed-Loop Liquid Cooling Does Not Mean Zero Water Consumption One of the biggest misconceptions about modern AI data centers is that liquid cooling automatically means enormous amounts of water are being consumed inside the buildings. In reality, many newer AI facilities use closed-loop liquid cooling systems that continuously circulate the same coolant through pipes connected to computer processors.

These systems transfer heat away from powerful graphics processing units, commonly known as GPUs, which perform the calculations needed to run artificial intelligence applications. Rather than relying entirely on fans to move hot air away from the equipment, liquid cooling brings coolant much closer to the processors, allowing heat to be removed more efficiently. The distinction is important because circulating water or another coolant through a sealed system is not the same as continuously consuming fresh water. A closed-loop system can operate with relatively little water replacement once it has been filled.

However, the cooling process does not end when heat leaves the computer chips. That heat must still be transferred outside the building and released into the surrounding environment. This second stage is where a significant portion of a data center’s water consumption can occur. Some gigawatt-scale AI data centers have been estimated to consume millions of gallons of water per day. However, that figure is not representative of every facility. Actual consumption depends on the cooling design, weather conditions, equipment and location. For example, Meta’s planned 1-gigawatt data center in Lebanon, Indiana, uses a closed-loop liquid cooling design. Water allocations for the project could reach 8 million gallons per day by 2031. The company has said this represents peak potential demand during the hottest conditions, rather than its expected everyday consumption.

The Two Main Ways Data Centers Release Heat

Once cooling systems capture heat from computer equipment, facilities must find a way to remove it. Two major approaches are dry cooling and evaporative cooling. Each comes with different demands for electricity and water.

Dry Cooling Uses More Electricity

Dry cooling systems use large fans and heat exchangers to transfer heat from inside a facility into the outside air. The technology works similarly to a radiator, allowing heated coolant to release its heat without relying on water evaporation. One advantage is that dry cooling can dramatically reduce the amount of water needed for daily operations. This makes it an attractive option in areas experiencing drought, water shortages or restrictions on municipal water supplies.

The trade-off is electricity consumption. When outdoor temperatures rise, fans must work harder to remove heat. Some systems also require mechanical refrigeration to lower temperatures enough to keep equipment operating safely. In particularly hot climates, these systems can become more energy-intensive because the outside air is less effective at carrying heat away. That creates additional electricity demand at a time when communities may already be using more power for residential air conditioning.

Evaporative Cooling Uses Water to Reduce Heat

Evaporative cooling takes a different approach. Instead of relying entirely on fans and refrigeration, these systems use water evaporation to carry heat away. Water circulates through cooling towers or across specially designed surfaces. As some of that water evaporates, it removes heat from the remaining water and helps lower the temperature of the cooling system. This process can be more energy-efficient than relying exclusively on mechanical air conditioning. However, the evaporated water must be replaced, creating a continuous demand for additional water.

Cooling towers also require water to be periodically drained and replenished to prevent minerals and other dissolved materials from accumulating. This process, known as blowdown, creates additional water demand.

Why Closed-Loop Cooling Does Not Eliminate Water Use

The difference between internal cooling and external heat removal is central to understanding the environmental footprint of AI infrastructure. A data center can use a closed-loop liquid cooling system to circulate coolant directly through its servers while relying on a separate cooling system to release the collected heat outdoors. In other words, the coolant circulating around the GPUs may be reused repeatedly, but the equipment responsible for releasing that heat can still consume water.

This is why statements that a data center uses closed-loop cooling or has zero water consumption need to be examined carefully. Those descriptions may refer to a particular part of the cooling system rather than the facility’s entire operation. Some facilities can use dry cooling to avoid routine evaporative water consumption. Others rely on cooling towers, while some combine different technologies depending on outdoor temperatures and operating conditions. The choice of cooling technology depends on factors including facility size, computing demand, location, climate and access to water.

How Much Water Do Data Centers Actually Consume?

There is no single water consumption figure that accurately represents every data center. A smaller facility handling traditional computing operations can have substantially different requirements from a massive AI campus filled with high-density GPU servers. According to estimates published by the U.S. Department of Energy and Lawrence Berkeley National Laboratory, U.S. data centers directly consumed approximately 17 billion gallons of water in 2023, compared with approximately 5.6 billion gallons in 2014.

Direct water consumption represents only part of the industry’s overall water footprint. Electricity generation can require additional water outside the data center itself, depending on the energy sources used to power the facility. The distinction between water withdrawal and water consumption is also important. Withdrawal refers to water taken from a source, while consumption refers to water that is not returned to the original source in an available form, often because it evaporates.

A facility can withdraw water, circulate it through its cooling infrastructure and return some of it to the water system. That does not mean all the water withdrawn has been consumed. Some data centers receive water through municipal systems, while others use reclaimed wastewater or obtain water from surface water and groundwater sources. That distinction matters to communities because a facility connected to a municipal water system may compete with other local demands, particularly during periods of drought or unusually high temperatures.

Data Centers Are Also Driving Electricity Demand

Water is only one part of the resource challenge. The electricity required to operate AI data centers is becoming an increasingly significant issue for the U.S. power grid. Every GPU, server, networking device and cooling system requires electricity. As AI models become more sophisticated and companies build larger computing facilities, the amount of power needed to operate these campuses continues to grow.

Research from Lawrence Berkeley National Laboratory estimates that U.S. data centers could account for approximately 11.8% of the country’s total electricity consumption by 2030. Projections vary depending on future data center growth and energy demand. Cooling is a significant part of that electricity demand. Fans, pumps, chillers and refrigeration equipment all require power to maintain safe operating temperatures.

This creates another engineering challenge. Facilities that reduce water consumption through dry cooling may require more electricity, particularly during periods of extreme heat. Facilities using evaporative cooling can reduce some of that electrical demand but consume more water. There is no universal cooling system that eliminates every resource concern. The best approach depends on the facility’s design, climate, electricity supply and available water resources.

Why Communities Are Fighting New AI Data Centers

The growing demand for electricity and water has turned data center development into a major community issue. Residents in several states have raised concerns about whether large computing facilities could place additional pressure on municipal water supplies, increase electricity demand, create noise pollution and affect household utility costs. On November 4, 2026, voters across the United States approved numerous ballot measures restricting or slowing data center development.

Ballot measure counts vary depending on which proposals are included. One reported tally identified 40 of 41 proposed restrictions passing, while a broader count identified 47 decided data center-related measures, with 46 outcomes favoring additional restrictions or requirements. The results reflect growing public attention to the local consequences of large-scale AI infrastructure.

In communities where water supplies are already under pressure, residents want to know whether data centers will rely on drinking water, reclaimed wastewater or other sources. They are also asking how much electricity these facilities will require and whether utility infrastructure improvements will ultimately affect local ratepayers. However, higher electricity demand does not automatically mean every household will experience higher bills. The impact depends on utility planning, infrastructure investments, electricity contracts, rate structures and whether developers pay their share of the costs.

Can Data Centers Operate Without Using Large Amounts of Water?

Yes, some data centers can significantly reduce their direct water consumption through alternative cooling technologies. Dry cooling, closed-loop liquid cooling and systems that use reclaimed wastewater can all help reduce demand for freshwater. Some operators are also exploring more efficient computer chips, improved heat exchangers and cooling designs that allow facilities to operate at higher temperatures without relying as heavily on mechanical refrigeration.

The challenge is that reducing water consumption does not necessarily eliminate the environmental footprint. A facility that relies on electricity-intensive cooling may shift some of its resource demand from local water supplies to the electrical grid. Location also matters. A cooling system that operates efficiently in a cooler climate may require substantially more electricity or supplemental cooling in a hotter region. For that reason, evaluating a proposed data center requires more than looking at a company’s advertised water consumption. Communities need information about both internal cooling and external heat rejection, as well as expected water consumption during the hottest days of the year.

The Bigger Question Behind the AI Boom

Artificial intelligence is expanding rapidly, and the infrastructure supporting it is becoming larger and more resource-intensive. Companies including OpenAI, Meta, Amazon, Google and other technology developers are investing in massive computing facilities to support increasingly demanding AI applications. But the physical infrastructure behind those services requires real resources. GPUs generate heat, cooling systems must remove it, and the electricity needed to keep everything running must come from somewhere.

Closed-loop liquid cooling is an important advancement because it allows data centers to cool powerful processors while recirculating coolant. However, it does not automatically solve the separate problem of releasing heat into the environment. The central issue is the trade-off between water and electricity. Evaporative cooling can conserve electricity while consuming water. Dry cooling can reduce water consumption while increasing electricity requirements, particularly in hot weather.

As more gigawatt-scale AI facilities are proposed across the United States, transparent reporting about water consumption, electricity demand and cooling technology will become increasingly important. For communities considering new data center developments, the question is not simply whether a company promises to use less water. It is how much water and electricity the entire facility will require, where those resources will come from and who will ultimately pay for the infrastructure needed to support it.