When Cornell University systems engineer Fengqi You began studying the environmental footprint of data centers three years ago, the artificial intelligence boom was still in its early stages. Yet even then, he and his research team noticed a glaring absence in the global discussion about AI: few were talking about how this rapidly expanding technology would impact energy, water, and other essential resources.
“When we started this, we saw that AI was growing very fast,” You said. “It was clear it would have to be aligned with power-grid planning, with water and other resource planning. There were no discussions about these topics—but we wanted to bring real numbers, rigorous analysis on AI’s physical footprints.”
This week, You and his team published their findings in the journal Nature Sustainability, revealing the staggering scale of AI’s potential environmental cost. Depending on how quickly the industry grows, U.S. data centers could consume as much water annually as 10 million Americans and emit as much carbon dioxide as 10 million cars. In other words, the environmental footprint of the nation’s AI infrastructure could soon rival that of the entire state of New York.
Despite these daunting projections, the researchers remain cautiously optimistic. They emphasize that proactive policy decisions—such as decarbonizing the power grid, increasing energy and water efficiency in data centers, and strategically locating new AI infrastructure—could significantly mitigate the damage. “We’re still early in this growth,” You said. “It’s in our hands now. It’s not too late yet to do the planning and accounting for resource constraints that could let AI continue to grow.”
The study also reveals stark regional disparities in how AI affects the environment. In particular, Northern Virginia’s “data center alley”, one of the world’s largest data center hubs, is facing severe limits on its energy and water supply. “Northern Virginia won’t have enough resources to support the growth of the AI industry in a sustainable way,” You warned. “There are too many constraints in terms of energy and water.”
Nevertheless, construction in Northern Virginia is expected to continue dominating the U.S. landscape through 2030. That is why You and his colleagues urge policymakers to start steering new projects toward other regions—especially Midwestern states such as Texas, Montana, and Nebraska, where renewable energy potential and water availability are far greater. “The key point we’re trying to make is that for the new projects, we have a chance to decide right now,” You said. “And we need to think about somewhere else.”
Redirecting AI infrastructure to more resource-rich areas could reduce the industry’s environmental toll without significantly slowing its growth. “States like Texas, Montana, Nebraska, for instance, have enough water supply and have good targets for getting clean energy and sufficient power supply,” You explained.
The findings arrive as global concern grows over AI’s accelerating energy and water demands. Utilities across the U.S. are already rushing to construct new gas-fired power plants to meet the surging power requirements of AI-driven data centers—raising fears that the technology’s expansion could undermine national emissions targets.
A recent report by the Center for Biological Diversity warned that if current trends continue, U.S. data centers could generate nearly half of all power sector emissions permitted under existing climate goals. The report added that because of AI’s fossil fuel dependency, every other sector of the economy would need to reduce emissions by an additional 60 percent to stay on track for the 2035 targets.
Environmental advocate Jean Su, a co-author of that report, cautioned that society must challenge the assumption that AI’s expansion is inevitable or inherently beneficial. “Technology optimists are saying AI is going to solve the climate emergency and cure cancer,” Su said. “But the way to actually resolve the climate emergency is to phase out fossil fuels. Scientists have already told us how to do it—we just need political will.”
That political will, however, appears increasingly uncertain. With President Donald Trump reversing renewable energy projects and promoting coal and gas, many experts worry that data centers will continue to rely heavily on fossil fuels. “Projections show that data centers are going to be powered by fracked gas through 2035,” Su said. “That’s in the political interest of the president and the fracked gas industry.”
Su and others stress the importance of public engagement and transparency in decision-making around data center construction. Communities, she said, should have a voice in determining whether such projects truly serve their interests.
Ultimately, You’s study underscores that the United States’ 2035 climate goals—eliminating carbon pollution from the power sector and slashing greenhouse gas emissions across the economy—remain achievable, but only if AI infrastructure is guided by strong policies. This includes setting AI-specific efficiency standards for energy and water use, similar to fuel economy benchmarks in transportation.
Both You and Su agree that the next few years will determine whether the AI revolution becomes a tool for sustainability or a new source of environmental strain. “We can’t let billionaire corporations dictate policy while the rest of us pay the price,” Su said. “We have to ask what’s actually in the public interest.”
source: inside climate news
