AI Companies Are Draining Water from Regions That Are Already Running Dry

Image generated with AI

Artificial intelligence requires an enormous amount of electricity to operate. Data centers that power large language models have to be cooled constantly, which requires a lot of water. By 2028, if current trends continue, AI infrastructure in the United States could consume as much water annually as 18.5 million households need for all indoor uses.

That is not a small number. In 2022, Google, Microsoft, and Meta alone used an estimated 580 billion gallons of water just to provide power and cooling to their data centers (Food and Water Watch, 2025).

The problem is where these data centers are located. More than 160 new AI data centers have sprung up across the United States in the past three years, and many of them sit in regions already facing water scarcity (Bloomberg, 2025). Arizona, Texas, and parts of the West have become prime locations for data centers because electricity is cheaper there and there is space to build. But these are also the parts of the country struggling with drought and competing demands for limited water.

The situation is even more stark internationally. In Saudi Arabia and the United Arab Emirates, companies are building data centers in some of the driest places on Earth. In China and India, an even greater proportion of new data centers are located in arid regions compared to the United States (Bloomberg, 2025).

What makes this particularly difficult is that when a data center uses water, it does not simply recycle it back to the source. Google's data centers, for example, only discharge 20% of the water they withdraw back to wastewater treatment. The other 80% is lost to evaporation (Food and Water Watch, 2025).

Tech companies made commitments to be water positive by 2030, meaning they would add more water to the environment than they use. But the AI boom has made those commitments increasingly difficult to keep.

There is also the issue of energy sources. Tech companies say they will power data centers with renewables, but current projections suggest that renewables will cover only 40% of new electricity demand from data centers through 2030. That means the rest will come from fossil fuels, and fossil fuel generation is far more water-intensive than renewable energy.

Chevron recently announced a partnership to build gas power plants specifically to supply energy to new AI data centers (Food and Water Watch, 2025).

The irony is that while AI is consuming massive amounts of water in drought-prone areas, AI itself could be used to predict and manage droughts. Machine learning models are becoming increasingly sophisticated at using satellite data, soil moisture, and weather patterns to forecast drought conditions with lead times that allow for real intervention (MIT, 2025).

In the UK, a team is using satellite technology and machine learning to detect water leaks in infrastructure, reducing detection costs and identifying problems faster than traditional methods (Illuminem, 2025).

AI can help optimize irrigation systems in agriculture by combining weather forecasts and crop data to deliver water more efficiently.

So we have a situation where the technology that could help us manage water scarcity more intelligently is simultaneously consuming water at unprecedented rates in precisely the regions that need to conserve it most.

Some researchers argue that locating data centers in regions with sustainable water supply and using closed-loop cooling technologies could reduce this pressure. But those solutions require investment and infrastructure that is not always available in the places where companies want to build

The bigger question is whether the benefits of AI development are worth the water cost when placed against local needs and climate resilience. When you are in a place like Arizona and there is a drought and your town is rationing water, it is hard to accept that a significant portion of your available water is cooling servers for a technology company a thousand miles away.