Down the street from Santa Clara University’s campus sits a nondescript, concrete block of a building, San Jose’s dry foothills glowing gold just to the east. Inside are stacks of computing servers, storage systems, networking infrastructure, and other hardware built to power artificial intelligence. It’s one of more than 50 such data centers in the city of Santa Clara, in the heart of Silicon Valley, one of the key places where this technology is being developed, refined, and delivered to the world.
All that data processing comes with a voracious appetite for energy. And it’s eating up U.S. cities’ power grids in many places. Locally in Santa Clara, about 60% of the city’s electricity is directed to these data centers, according to officials with the city and Silicon Valley Power. This is not a surprise.
The increasing carbon footprint of generative artificial intelligence has received a fair amount of recent attention. What hasn’t been as widely discussed, yet, is AI’s water consumption. Data centers generate a lot of heat, so copious amounts of water may be pulled from communities for AI cooling. And there’s increasing concern that vulnerable communities—whether they are of lower socioeconomic status or already dealing with depleting water resources—will be impacted more than others.
Iris Stewart-Frey, professor in the Department of Environmental Studies and Sciences, and Irina Raicu, director of Internet Ethics at Markkula Center for Applied Ethics want to talk about it more. The pair received a $50,000 grant to study how water use by AI data centers impacts water availability and distribution from the Next 10 Foundation, a nonprofit focused on the intersection of the economy, environment, and quality of life issues in the state of California.
A hydrologist by training, who runs Santa Clara University’s Water and Climate Justice Lab, Stewart-Frey says California is a fitting setting to study this issue. “Much of California is semi-arid. A lot of its water is already spoken for, and many of its basins are overdrafted. Under climate change, we expect to be subject to more frequent droughts,” she says. Indeed, this past June was the 11th driest June on record in California since tracking began in 1895. “We have the technology boom in Silicon Valley, but just an hour away [in the Central Valley’s agricultural basin], people are drawing drinking water from shallow, polluted wells.”
Stewart-Frey and Raicu linked up on this issue last fall when the Markkula Center sponsored a conference about AI’s impact on the environment at Santa Clara. A year prior, Raicu had written an op-ed in the San Francisco Chronicle calling for more focus on the ramifications of the artificial intelligence revolution on the environment. In her research for that article, Raicu learned that just one “conversation” with ChatGPT of 20-50 prompts and replies uses the equivalent of a standard disposable water bottle. (Such estimates are often challenged and can vary depending on many factors.). “So much of the push to integrate AI, and especially generative AI, into every aspect of our lives has been about highlighting all its benefits,” says Raicu. “And there are many benefits. But people have not been told about the drawbacks, and this is a massive one that I think most people are still not aware of.”
Part of the problem, Raicu says, is that the private companies building these AI data centers or renting space in them are not publicly releasing data about their power and water usage. “You can’t get answers,” she says. “So it’s hard. And with everything going on in the world, people already have so many things on their minds that water for data centers is kind of far down the list. But it’s going to be especially far down the list if you can’t get the numbers to put in front of them or in front of local government entities to tell them, ‘Hey, you actually have to do something about this.’”
While the specifics of the study are evolving with the data they are analyzing, Stewart-Frey and Raicu plan to use spatial analysis and interviews, and work with students to build case studies of water usage of several data centers, both existing and under development. Stewart-Frey hopes they’ll find enough information to extrapolate across the whole of California, but much of that depends, of course, on access. “We’ll look at environmental impact reports, what water supply is available, the history of how water is divided in the area. We’ll look at the socioeconomic vulnerability of the local community, if there’s been any concern or opposition, or any knowledge about the data centers at all,” she says.
Like many universities across the country, Santa Clara has been looking for ways to incorporate artificial intelligence into its campus and curricula. In June, the School of Engineering announced the launch of a new master of science in artificial intelligence program and plans to welcome its first cohort this fall. Much of the emphasis on AI at SCU is on the ethical responsibility of its creators and real-world applications. Stewart-Frey and Raicu see their work not as separate from that charter but very much in line with Santa Clara’s ethos to build leaders of conscience, competence, and compassion.
“If you don’t understand both the positive and the negative aspects of a technology, you cannot be a competent developer or user of it,” Raicu says. “So it’s absolutely tied into what we are trying to do at Santa Clara around AI. … And I think as a Jesuit university, we’re absolutely called to raise all these issues.”