A single AI data center could use as much water as a small town, raising questions in North Carolina about growth, climate resilience and resource planning.
Posted
By Liz McLaughlin, WRAL climate change reporter
The rapid growth of artificial intelligence is fueling demand not just for electricity but also for water, raising new questions for communities in North Carolina.
Industry estimates show a 250-megawatt data center can use about a billion gallons of water each year, or 2.7 million gallons a day.
Other WRAL Top Stories
Ed Buchan, assistant director of Raleigh Water, said that level of demand would be significant for the region.
“It’s a meaningful chunk of water that would otherwise be reserved for residential or commercial growth,” said Ed Buchan, assistant director of Raleigh Water. “That’s about what the demand is for some of our merger communities like Knightdale and Wendell.”
Buchan said that most water utilities return nearly all of what they treat back to the river basin, but cooling systems at data centers often evaporate it. “If it’s evaporated, it’s basically out of the equation,” he said. That could pose risks during drought, when downstream flow is critical.
Shaolei Ren, a University of California Riverside professor who has studied AI’s water footprint, said usage can spike during the hottest months. That means competition with residents who already need more water in summer. “Understanding the peak usage is also important in addition to the annual total,” Ren said.
While the overall national water use of data centers remains small compared with agriculture or industry, Ren noted that local impacts can be far greater. “We need to look at the specific context and understand if it is creating some pressure on the water infrastructures,” he said.
Michael Pencina, Duke University’s vice dean for data science, said sustainability must be part of AI’s future. At Duke, researchers are exploring ways to recycle the heat generated by computing power. “We’ve looked at using the heat from GPU farms to warm up water that’s used across campus,” he said. “It’s about being creative in reducing the footprint while still advancing discovery.”
AI companies have pledged to become “water positive” by 2030, yet transparency remains a challenge. Ren said roughly half of data centers are not publicly disclosing their water use. “Without knowing the numbers, how can we manage these resources responsibly,” he asked.
A Washington Post analysis found a ChatGPT text query uses only about five drops of water, but those small amounts add up when multiplied by billions of requests and the massive cooling demands of data centers.
Buchan said Raleigh has enough supply to handle one or two large facilities, but multiple sites could speed the need for new reservoirs or other costly projects. “If you had a significant number of them come into your service area, and then they were the type that evaporated a lot of water, then that might move up your need for a new water resource pretty quickly,” he said.
Communities weighing data center proposals are often drawn to their high property tax value.
“Data centers clearly have benefits,” Ren said. “On the other hand, they also use resources like power and water. The real question is whether those tradeoffs add up to a net benefit for the community.”
- Related
artificial intelligence climate change data mining water
© 2025 Copyright Capitol Broadcasting Company
