AI not only consumes electricity but also water

Wallstreetcn
2025.07.16 03:21
portai
I'm PortAI, I can summarize articles.

Thousands of data centers in the United States require massive amounts of water for cooling, which is particularly problematic in drought-stricken states. Traditional large data centers can consume as much water in a day as a town of tens of thousands of people, and generating AI dialogues also requires a significant amount of water. In addition to exacerbating water resource shortages, the chemical additives in cooling systems pose long-term risks of groundwater pollution. Tech giants like Amazon and Google are actively taking measures, such as using recycled water and developing waterless cooling technologies

As people are hotly discussing the surge in electricity consumption brought about by artificial intelligence, a relatively hidden but equally severe resource challenge is emerging—water scarcity.

With the rapid development of AI technology, the construction boom of data centers hides a significant water consumption crisis. The latest research from market intelligence group Aterio shows that there are currently 1,827 active data centers in the United States, with another 1,726 announced construction plans and 419 under construction.

These data centers require millions of gallons of water for cooling, and their water demand is growing in tandem with the expansion of AI-supported facilities. More concerning is that multiple data centers are being planned or constructed in drought-stricken areas facing water shortages. In states like Nevada, Arizona, Texas, Utah, California, and Colorado, where water resources are severely strained, there are 437 data centers either under construction or planned.

Environmental experts warn that, in addition to the massive water usage, the chemical additives in data center cooling systems may also contaminate groundwater, causing long-term environmental impacts.

AI Water Consumption is Astonishing

Steffen Lehmann, a professor of architecture and urban planning at the University of Nevada, Las Vegas, points out that traditional large data centers can consume up to 5 million gallons of water per day, equivalent to the daily water usage of a town with a population of 20,000 to 50,000.

According to a joint study by the University of California, Riverside, and the University of Texas at Arlington, even relatively simple AI workloads require a significant amount of cooling water.

For example, generating 10 to 50 medium-length responses with ChatGPT consumes the equivalent of a 500-milliliter bottle of water. According to Semrush, ChatGPT has approximately 5.24 billion visits per month.

McKinsey & Company’s analysis predicts that by 2030, the demand for AI-enabled facilities will grow at a rate of 33% per year.

Pollution Risks Cannot Be Ignored

Steve Rosas, president of environmental services company Omega Environmental Services, states that data centers face not only water usage issues but also the risk of cooling system additives leaching into groundwater and causing contamination.

"We have remediated sites where industrial cooling operations used biocides, corrosion inhibitors, and scale inhibitors that contaminated soil and groundwater; these chemicals can persist in the environment long after the facilities are shut down."

He is particularly concerned about PFAS (per- and polyfluoroalkyl substances) contamination. These substances, known as "forever chemicals," are named for their persistence in the environment. Rosas states:

"Many components of cooling systems and firefighting foams contain PFAS chemicals, and we are currently investigating these substances at over 600 water source locations in California."

Arnold Pinkhasov, a software engineer at tech accelerator OSLabs, adds that large-scale water extraction may lower groundwater levels, threaten local wildlife, and even compete with agricultural water usage. When the water used for cooling returns to the environment at higher temperatures, it can also cause thermal pollution, affecting river and lake ecosystems

Regulatory System Has Gaps

In Virginia, which has the largest data center market in the world, there are currently no statewide regulatory provisions regarding water usage for data centers. The state has over 150 data centers just in Northern Virginia, highlighting the contradiction between regulatory needs and development progress.

Virginia House Bill 1601 was supposed to require environmental impact assessments for surface water and groundwater for proposed data center facilities, but the bill was vetoed by Governor Glenn Youngkin in May. Youngkin stated that the legislation would limit local autonomy and create "unnecessary red tape" for new data centers.

Jon Hukill, communications director of the Data Center Alliance, stated that the industry as a whole is committed to responsible water usage but faces challenges in a legislative environment where regulations vary significantly from state to state.

Tech Giants Actively Respond

Amazon Web Services (AWS) stated that it is "doubling down on protecting freshwater resources," aiming to achieve "net water benefits" in its data centers by 2030, meaning the amount of water returned to the community exceeds its direct operational usage. An AWS spokesperson revealed to the media that the company focuses on using more sustainable water sources, such as water recycling or rainwater harvesting.

The spokesperson disclosed that AWS has achieved 53% of its goal for net water benefits by 2030, up from 41% in 2023.

In the Americas, our existing data centers do not require water for cooling 90% of the time. Additionally, our water replenishment projects in the U.S. and globally are expected to replenish over 9 billion liters of water annually for the environment and communities.

Google uses recycled or non-potable water in over 25% of its data centers. Microsoft launched a new design last August that does not require water to cool AI-related workloads in its facilities