AI power use forecast finds the industry far off track to net zero

As the artificial intelligence industry rapidly expands, questions about environmental impact of data centers are coming to the fore – and a new forecast warns the industry is unlikely to meet net zero targets by 2030.

Fengqi Yu from Cornell University in New York and his colleagues modeled how much energy, water and carbon today's leading artificial intelligence servers could use by 2030, taking into account various growth scenarios and possible data center locations in the United States. They combined projected chip supplies, server power consumption and cooling efficiency using state-by-state grid data for analysis. While not every AI company has set a net zero target, some larger tech companies active in AI, such as Google, Microsoft, and Meta, have set goals with a 2030 deadline.

“The rapid growth of AI computing is essentially changing everything,” says Yu. “We're trying to understand how the growth of the sector will affect this?”

They estimate that by 2030, building artificial intelligence servers in the United States will require between 731 million and 1.125 billion additional cubic meters of water, and emissions will be equivalent to between 24 and 44 million tons of carbon dioxide per year. The forecast depends on how quickly demand for AI grows, how many high-performance servers can actually be built, and where new data centers are located in the US.

The researchers modeled five scenarios based on growth rates and identified different ways to reduce the impact. “Number one is location, location, location,” Yu says. Locating data centers in Midwestern states, where water is more accessible and the grid is powered by a higher share of renewables, can reduce the impact. The team also believes that decarbonizing the energy supply and improving the efficiency of data center computing and cooling processes are key ways to limit the impact. Together, these three approaches could reduce the industry's emissions by 73 percent and its water footprint by 86 percent.

But the group's forecasts could also be undermined by public opposition to the installation of data centers due to their potentially harmful environmental impacts. In Virginia, where there is about one eighth of the world's data center power, residents have began to put forward opposition to further planned construction, citing impacts on their water supplies and the environment in general. Similar petitions against data centers have been filed in Pennsylvania, Texas, Arizona, California and Oregon. Data from Data Center Watch, a research company that tracks data center development, suggests local opposition is deadlocked. Projects worth $64 billion.. However, even in places that have successfully abandoned data centers, it is unclear how much energy and water they can use.

That's why the new findings have been welcomed—albeit cautiously—by those trying to study and quantify the environmental impacts of AI. “AI is such a rapidly evolving field that it is very difficult to make any meaningful predictions about the future,” says Sasha Luccioni at AI company Hugging Face. “As the authors themselves say, industry breakthroughs could fundamentally change compute and power requirements, like what we saw with DeepSeek,” which used various techniques to reduce brute force computation.

Chris Preist at the University of Bristol in the UK say, “the authors are right to point out the need to invest in more renewable energy capacity,” and add that data center siting issues are important. “I think their assumptions about using water to directly cool AI data centers are pretty pessimistic,” he says, suggesting that the model's “best case scenario” is more like “business as usual” for data centers these days.

Luccioni believes the document highlights what is missing in the world of artificial intelligence: “more transparency.” She explains that this can be remedied by “requiring modelers to track and report their computing and energy use, and make this information available to users and policymakers, and make strong commitments to reducing their overall environmental impact, including emissions.”