Tech companies have invested so much money in building data centers in recent months, it’s actively driving the US economy—and the AI race is showing no signs of slowing down. Meta chief Mark Zuckerberg told President Donald Trump last week that the company would spend $600 billion on US infrastructure—including data centers—by 2028, while OpenAI has committed already to spending $1.4 trillion.

An extensive new analysis looks at the environmental footprint of data centers in the US to get a handle on what, exactly, the country might be facing as this buildout continues over the next few years—and where the US should be building data centers to avoid the most harmful environmental impacts.

The study, published in the journal Nature Communications on Monday, uses a variety of data, including demand for AI chips and information on state electricity and water scarcity, to project the potential environmental impacts of future data centers through the end of the decade. The study models a number of different possible scenarios on how data centers could affect the US and the planet—and cautions that tech companies’ net zero promises aren’t likely to hold up against the energy and water needs of the massive facilities they’re building.

Fengqi You, a professor in energy systems engineering at Cornell and one of the authors of the analysis, says that the study, which began three years ago, comes at “a perfect time to understand how AI is making an impact on climate systems and water usage and consumption.”

The AI industry “is growing much faster than we expected,” he adds—especially with the Trump administration’s laser focus on the industry. “This whole thing is just getting so much momentum right now.”

Not all data centers are created environmentally equal: a lot of their water and carbon footprint depends on where they’re located. Some US states may have grids that run more on renewable energy, or are making big strides in putting more clean energy on the grid; this greatly lessens the carbon emissions from data centers that draw power from those grids. Similarly, states with less water scarcity are better suited to provide the large amounts of water needed for cooling data centers. (Cooling also constitutes a big part of data center energy use.) The best locations for a data center over the next few years in the US are states that strike a balance between these two inputs: Texas, Montana, Nebraska, and South Dakota, the analysis finds, are “optimal candidates for AI server installations.”

Much of the data center buildout in the US has historically focused on places like Virginia, the data center hub of the US, and Northern California. Being close to Washington, DC, and Silicon Valley was important to data center companies, as were the dense fiber connectivity in those regions and their skilled workforces. Virginia has also offered substantial tax breaks for data centers for years—one technique other states are turning to to lure development. According to Data Center Map, an industry tool that tracks data center development, of the 4,000-plus data centers in the US, more than 650 are in Virginia—the most in the country—and California has more than 320, ranking third.

While Virginia doesn’t suffer from water scarcity, advocates have said that massive energy requirements from data centers could derail the state’s goals to source 100 percent clean energy by 2045. Conversely, California’s long-running water issues may cause a problem if data centers keep expanding there. (In October, Governor Gavin Newsom vetoed a bill that would have required data center operators to disclose how much water they use, claiming that the state is “well positioned to support the development” of data centers.)

Data center operators choose locations based on a variety of factors, including energy and water needs. But they’re not the only inputs: Arizona, which has more than 160 data centers, is named in the analysis as one of the states facing “severe water scarcity issues.”

The industry has no plans to move out of some of their favored areas: There are still massive projects planned for Virginia, including a $9 billion investment from Google announced in August. But continue to build in already-stressed areas and “you’re going to exceed the capacity for natural resources,” You says. He compares data center operators flocking to popular areas as analogous to people living in the same area all going to the same grocery store at the same time. “The shopping experience is not going to be pleasant, even if it may be the best shop in town.”

Texas, one of the top states for future development named in the study, has built out a robust data center industry, and is now the second-most-popular state for data centers in the country, according to Data Center Map. (Much of this boom has come very recently: Data center construction in Texas quadrupled between 2023 and 2024.) But other top states named in the analysis—Montana, Nebraska, and South Dakota—still have relatively few data centers. However, their numbers are rising, You says, as the data center industry aggressively expands across the country. An LA-based company is looking to build the first hyperscale data center in South Dakota, while Nebraska’s 39 data centers include recently opened facilities owned by Meta and Google. Still, these states have a long way to go to catch up to places like Virginia. (South Dakota, according to Data Center Map, has just five facilities, the second-lowest in the country.)

As with any studies about AI, making predictions about the future is tricky, no matter where data centers end up getting built. The study acknowledges that a number of factors—from improvements in model efficiency, to advances in cooling technology, to changes in what kinds of energy gets added to the grid—could drastically change the amount of energy and water that ends up being used over the next few years. There’s also a possibility that the AI bubble may burst, leaving a scattered collection of half-built projects and contracts around the country. (Power issues are already leaving some data centers sitting idle for years.)

Part of what could happen depends on the political will for moving off fossil fuels. Nebraska, for instance, has massive potential for wind energy—one of the reasons it’s named in the analysis as a good spot for data centers—but has to date not actually built out that capacity; Nebraska’s utilities have instead invested heavily into more natural gas this year. The Trump administration, meanwhile, has spent months putting its finger on the scale to promote fossil fuels to power the AI boom.

A grid that doesn’t transition to renewable energy could mean serious trouble for the climate, given how aggressively data centers are expanding across the country. In the most extreme scenarios, the analysis finds, the US’s data center buildout could generate up to an extra 44 million tons of carbon dioxide equivalent each year—more than entire countries like Hungary, Portugal, and New Zealand each generated in 2022.

The worst-case scenario for the environment, You says, is if AI demand outstrips efficiency gains in computing in the coming years, while the transition to renewable energy slows down. “We know demand is picking up very quickly, but it could be even faster with all these new applications, more devices, agentic AI, all these things,” says You. “It’s going to be a problem.”

But the study also claims that developments in technology, like those for cooling and powering data centers, in addition to siting data centers correctly, could drastically improve both emissions and water use. Outside experts, meanwhile, caution against trying to predict emissions from what is still a rapidly evolving industry, with lots of different variables involved.

“I try not to put too much value on specific numbers,” says Noman Bashir, the Computing and Climate Impact Fellow at MIT’s Climate and Sustainability Consortium, who was not involved in the study. Bashir points out that data centers installing their own energy onsite—like building their own natural gas plants or installing solar panels and batteries separate from the grid—as well as advancements in some key new technologies, like nuclear, could have a much bigger impact on overall emissions than the study calculates.

But Bashir praises a base conclusion of the paper: that tech companies that made net-zero pledges are unlikely to meet them, given the rush to build data centers. Big Tech players like Google and Microsoft have quietly acknowledged in recent sustainability reports that their focus on AI is making the promises they made on cutting emissions much harder to achieve.

You hopes that his work helps create more transparency on emissions from the companies driving the buildout—something, he says, that could be analogous to the nutrition labels on food.

“The future is going to be much faster, much bigger,” he says. “Keeping sustainability in mind in the early stage is much better than later on, in terms of how AI computing infrastructure will develop and grow in our country.”