AI Water Usage: Tackling Data Center Water Strain

Artificial intelligence is hot, and it gets headlines, investments and users. It is also operated literally, literally: Data centers models that operate artificial intelligence (AI) use large amounts of electricity and generate huge heat. To keep the servers from high temperature, many facilities depend on the cooling systems that use water.

The use of artificial intelligence data centers comes in two forms. Outside the water that cools the servers, databases indirectly contribute to the use of water by generating the electricity needed to operate its operations. This indirect use often constitutes 80 percent or more of the total water use.

Reducing the water fingerprint of artificial intelligence means treating two very different problems – what happens inside the walls of the data center, and what is happening behind it on the energy network.

Live water use: local and sometimes exhausting

Just as human bodies themselves cool by sweating, data centers are often cooled by water evaporation – a process that dispels the heat and leads to water loss in the atmosphere, and thus is calculated as “consuming”. In many cases, water is withdrawn from the same municipal systems that provide homes and companies.

While most major technology companies now reveal their direct use of water, not all data centers follow, which makes the overall image unclear. In recent reports, companies estimated that between 45 percent and 60 percent of the withdrawn water is consumed.

According to a recent report by the Lawrence Berkeley National Laboratory, direct water consumption for 2023 by databases in the United States – which is approximately 40 percent of the world databases – is estimated at 17.5 billion gallons. Assuming the consumption rate by 50 percent, this means 35 billion gallons of water withdrawal, or about 0.3 percent of the total general water supply of the United States. The same report shows that water consumption in the American data center can double or even quadruple level 2023 by 2028.

At the national level, the use of data centers water is relatively modest. But in some areas where data centers are concentrated – especially in areas that already face a shortage – pressure on local water systems can be large. Bloomberg News Reports indicate that about two -thirds of the American databases designed since 2022 are in high areas of water.

In Newton Province, Georgia, according to what was reported, some proposed data centers requested more than a day than the entire province uses daily. Officials are now facing difficult options: rejecting new projects, or they require water -saving alternative cooling systems, investing in costly infrastructure promotions, or risks that impose water legalization on the population.

The greatest stress may not be the total, but timing. On hot days when residents and companies need water more than that, the water demand in the data center increases as well. In Arizona, the use of monthly water in the data center during the summer can be twice the level of average.

Indirect use of water: thirsty electricity

The other part of the equation is electricity that operates databases. In many places, electricity-whether to train artificial intelligence models in databases or operate a lamp at home-by powerful fossil-based power plants require their cooling water. The electric power sector in the United States pulls about 11.6 gallons of water and consumes 1.2 gallons per kilowatt hours of electricity produced, which places it among the largest water users in the country. Water used to produce electricity that operates data centers is indirect use of water.

The water used by power plants is usually inaccurate and not drawn from municipal water systems. However, tension can be placed on rivers, water classes, and ecosystems-especially in water control areas.

For most American data centers, this indirect use is much higher than the use of direct water on the site. One of the paper was estimated that in 2023, using GPT-3 to create one text output ranging from 150 to 300 words consumed a total of 16.9 milliliters of water in an average American data center -2.2 ml for cooling on site and 14.7 ml to generate electricity. It is possible that the gains of efficiency in subsequent models have reduced these numbers, but the use of indirect water still prevails.

How to reduce the effect of data centers water

Unlike electricity, the data center cooling systems is to choose a design. Low -cost and effective flying cooling, but it can inhibit local supplies during the summer heat waves, when the water is the most needed and less needed. To manage this peak demand, data centers can build water storage on site or install thermal energy storage. Water infrastructure upgrade – such as expanding distribution or leaks repair – also can also help local systems help in dealing with demand mutations.

The alternatives to the evaporative cooling include a liquid and liquid cooling, using recycled water to cut the use of drinking water, and re -use waste temperature to reduce the demand for cooling. Some advanced designs for refrigeration water recycling in a closed loop, so water is not consumed; These “zero-water” designs eliminate the need to benefit from local drinking water supplies. However, many of these designs raise the demand for electricity, which in turn can increase the indirect use of water. Water refrigerated data centers consume about 10 percent of less energy than the refrigerated data centers.

In indulging cooling systems, the servers are immersed in a liquid that carries heat away without evaporation of water. Jason Aldeen/Bloomberg/Getty Emoz

In the stressful areas of water, the priority should be low to water cooling systems to zero to reduce direct use, while investing to add renewable energy sources to local networks to reduce indirect water use and reduce carbon emissions from high demand for electricity. In wet areas with carbon dense networks, priority should be given to reduce energy use to reduce total water consumption, even if this means continuing to use evaporative cooling with high water consumption.

The reality of interlocking water and electricity systems forces data data operators to move in difficult differentials between global climate goals and local water needs. Often these options are not simple, but until renewable energy sources dominate electricity networks, they may be inevitable.

The opinions expressed in this article are the opinions of the authors and do not necessarily reflect the views of employers or affiliated institutions.