Why AI needs so much water

Image: Illustration / Digiopedia

Artificial intelligence is often discussed in terms of computing power and electricity. But there is another resource behind the rapid expansion of AI infrastructure: water.

AI models run in data centers filled with high-performance processors. Those chips generate substantial heat, and keeping them within safe operating temperatures requires cooling systems. Depending on how a facility is designed, that cooling can require significant amounts of water.

The scale of the impact varies widely, however. Water use depends on the cooling technology, local climate, server efficiency, electricity source and even how heavily the equipment is being used. A 2025 study from Lawrence Berkeley National Laboratory found that water use per data-center workload can vary by more than 10,000 times depending on these factors. 

Why data centers need water

Water is primarily used to remove heat from computing equipment.

Traditional cooling systems can use evaporation to carry heat away from a data center. As water evaporates, it takes heat with it, helping keep servers operating at the required temperature.

This becomes particularly relevant for AI infrastructure because modern AI accelerators can operate at very high power densities. More powerful processors generate more heat, increasing the demands placed on cooling systems. Microsoft, for example, says advanced AI chips require more intensive cooling because of the additional heat they produce. 

But not every data center uses the same approach. Some facilities rely heavily on outside air, while others use chilled water, evaporative cooling or liquid cooling systems.

There isn't one number for AI's water use

One of the biggest problems with broad claims about AI water consumption is that there is no universal amount of water associated with an AI request.

A workload's water footprint can change dramatically depending on where it runs and how that data center is operated. Lawrence Berkeley National Laboratory's research identified server efficiency, the water intensity of the electricity grid, cooling technology, climate and infrastructure efficiency among the major factors determining water consumption. 

That means an AI workload running in one data center could have a very different water footprint from the same workload running somewhere else.

Even estimates for individual AI queries should therefore be treated carefully. Google's research into the environmental impact of AI inference, for example, notes that calculations can change substantially depending on what parts of the supporting infrastructure are included. 

Water isn't only used directly

The water footprint of AI does not necessarily stop at the data center.

Electricity generation can also require water, meaning an AI workload can have an indirect water footprint associated with the electricity used to operate its servers.

Research examining AI's environmental impact therefore considers both direct water consumption from data-center cooling and indirect water use associated with energy production. 

This makes the overall picture more complicated than simply measuring how much water flows through a cooling system.

Location can make the difference

Water use becomes a more significant concern when large data centers are built in areas where freshwater supplies are already under pressure.

A facility using substantial amounts of water may have a relatively small impact on a region with abundant supplies, while the same facility could create greater competition for water in a drought-prone or water-stressed area.

Recent developments show why location is becoming part of the discussion around AI infrastructure. In India, for example, a large proposed Google data-center project has faced opposition related to potential pressure on local water resources, while Google says it plans to use advanced cooling technologies to reduce its impact. 

The issue is therefore not simply how much water AI uses globally, but where that water is being used and whether local supplies can support the additional demand.

The industry is trying to reduce water use

Data-center operators are developing cooling systems designed to reduce or eliminate the need for fresh water.

Microsoft introduced a new data-center design in 2024 that uses closed-loop, direct-to-chip cooling and does not require water evaporation for cooling. The company says the design can avoid more than 125 million liters of water per year per data center compared with its previous approach. 

AWS is also developing cooling technologies aimed at reducing water consumption in high-density AI infrastructure. Its in-row heat exchanger system is designed to capture heat directly from AI hardware and reduce how often water-based cooling is required. 

Other approaches include using reclaimed or recycled water instead of potable water, improving cooling efficiency and designing facilities around local climate conditions. 

There can be trade-offs, though. Microsoft notes that replacing evaporative cooling with mechanical cooling can increase electricity use, meaning reducing water consumption can sometimes require additional energy. 

AI's water footprint is likely to grow with demand

The rapid expansion of AI infrastructure means water use is becoming an increasingly important consideration for data-center planning.

A study published in Nature Sustainability estimated that AI-server deployment in the United States could result in an annual water footprint of 731 million to 1.125 billion cubic meters between 2024 and 2030, depending on the scale of deployment. The study also emphasizes the uncertainty created by differences in where and how AI servers are deployed. 

Those figures should not be interpreted as water consumed by AI alone in every location. They are projections based on specific deployment scenarios and include the broader water footprint associated with AI servers.

A problem of scale, efficiency and location

AI's water use is not simply a story about computers “drinking” enormous amounts of water. It is a combination of heat, cooling technology, electricity generation, infrastructure and location.

At the same time, the industry is improving efficiency and developing systems that can use little or no water for operational cooling.

As AI models become more capable and data centers continue to expand, the challenge will be balancing computing growth with the availability of electricity and water — particularly in regions where those resources are already under pressure.

For AI, water is another infrastructure question alongside chips, electricity and data-center capacity. How significant that question becomes will depend largely on where new facilities are built and how efficiently they use the resources available to them.