Adithi Ramakrishnan, Associated Press
Linley Sanders, Associated Press
On the way home from a recent trip, Eriech Tapia drove past a sea of dark industrial buildings lining an already jam-packed northern Virginia road: data centers.
"It was interesting to see how many were going up," said Tapia, a 30-year-old Republican from Arlington, Virginia, who works in higher education advocacy and policy. He said the data centers lend valuable computing power that can help keep the U.S. competitive with other nations, but he also worries about how much electricity they use.
According to a new poll from The Associated Press-NORC Center for Public Affairs Research and the Energy Policy Institute at the University of Chicago, Americans have grown more concerned about the environmental impacts of artificial intelligence over the last year. The survey data indicate that about half, 53%, of Americans are "extremely" or "very" concerned about artificial intelligence's environmental impacts, up from 41% last year. While Democrats are driving much of the increase in environmental concern, the poll also reveals broader worries about the local impact of data centers, which power AI and cloud computing.
Michael Greenstone, who directs the University of Chicago institute that collaborated on the survey, stated, "We're undertaking a massive experiment in real time, and I don't think there'll be any putting the milk back in the bottle. Change is scary, and I think some of that is being reflected here."
There are thousands of data centers in the U.S. with construction planned for more. These include smaller, conventional centers that have been around for decades, as well as larger hyperscale ones that are being built to serve artificial intelligence's added processing needs. Virginia and Texas have the most facilities currently in operation, with California, Ohio, and New York also being significant locations.
While 87% of existing data centers are in urban areas, most new data centers are planned for rural areas, according to a Pew Research Center analysis from earlier this year.
Nancy Antunes, a 60-year-old retired Democrat from Waltham, Massachusetts, expressed concern, stating, "I think that decisions are being made for the company, not for the community that they're being built in."
A recent United Nations University report predicted that data centers' environmental footprint, which already rivals some of the world's largest countries, will grow rapidly as AI is more widely adopted. The poll found that about two-thirds of Democrats say they are "extremely" or "very" concerned about how AI will affect the environment, up from about half in 2025. By contrast, about half of independents and about 4 in 10 Republicans are highly concerned about the environmental impact of AI.
Overall, Americans are more worried about the environmental impact of AI than other industries with a substantial environmental footprint. Only about 3 in 10 U.S. adults are highly concerned about the impact of meat production or air travel.
While concerns about environmental impacts are higher among Democrats, there is a broad sense of wariness about a rapid data center boom. About 6 in 10 Americans support limiting the number of new data centers that can be built, including most Democrats and Republicans.
Gedeon Kebede, a 21-year-old computer science student and Democrat from Cincinnati, Ohio, noted, "There's a lot of water usage by AI data centers that's taking away from other things like agriculture and farmers and places that really need it, and people who really need it."
A majority of Americans are "extremely" or "very" concerned about the impact of data centers on electricity prices or the water supply in the communities where they are built. Fengqi You, an energy systems engineer with Cornell University, stated that data centers' use of water to cool AI chips and computers is "nationally modest but potentially locally decisive."
Data centers are powered by tremendous amounts of electricity, particularly larger ones, raising concerns about where that energy will come from. According to a Lawrence Berkeley National Laboratory report, data centers could account for 11.8% of the U.S.' electricity consumption by 2030. Erik Nordman, director of Michigan State University's Institute of Public Utilities, mentioned, "We're talking about building electric infrastructure on a scale of entire cities, or in some cases, entire states just to serve one data center."
About two-thirds of U.S. adults support requiring data centers to use clean energy sources, such as wind or solar, to supply their electricity needs. This includes about 8 in 10 Democrats and 56% of Republicans.
However, some Americans feel that data center worries are distant. Mary Fisher, 68, a retired Republican from Jacksonville, Florida, stated, "You know, it hasn't come across my desk as something I should be concerned about."
About one-quarter of U.S. adults say data centers are "extremely" or "very" beneficial when it comes to technological advancement, and about 2 in 10 say the same about job creation. Many U.S. adults say data centers are "somewhat" beneficial or aren't sure, while about one-quarter say they're "not very" or "not at all" beneficial for technological advancement, and about a third don't see a benefit for job creation.
Tapia noted, "I see a huge increase in the number of temporary construction workers that are coming to areas, which is vitally important to the U.S. economy."
Republicans are more likely than Democrats to see data centers' benefits, but about 3 in 10 Republicans still think the centers are "not very" or "not at all" beneficial for improved infrastructure and job creation.
Kebede expressed his feelings about the pressure to use artificial intelligence, stating, "I think a lot of people just feel that it's very much a bubble. Ultimately, when it pops, it's going to be very detrimental to everyone, whether you bought into the AI hype or not."
The AP-NORC poll of 3,424 adults was conducted July 6-24, 2026, using a combined sample of interviews from NORC's probability-based AmeriSpeak Panel, which is designed to be representative of the U.S. population, and interviews from opt-in online panels. The margin of sampling error for all respondents is plus or minus 2.2 percentage points. The AmeriSpeak panel is recruited randomly using address-based sampling methods, and respondents later were interviewed online or by phone. To incorporate the nonprobability sample, NORC used TrueNorth calibration, an innovative hybrid calibration approach developed at NORC to explicitly account for potential bias associated with the nonprobability sample.