Author :
Lauren Leffer
Category :

How the AI Boom Threatens to Short-Circuit Climate Action

 Audubon > News Read More 

Northern Virginia’s Data Center Alley lives up to its nickname. Vast, windowless buildings proliferate beside suburban neighborhoods, looming over cul-de-sacs. Dwelling among these imposing buildings—warehouses full of servers, the physical embodiment of the online world—feels like “living inside a computer,” says Julie Bolthouse, director of land use at the nonprofit Piedmont Environmental Council. 

The area has been a hub for digital infrastructure for decades, but recently, development has exploded. In 2016 Loudoun County hosted 8.8 million square feet of data-center floor space. By 2020 that figure had more than doubled, and it has since doubled once again, with yet another doubling on the horizon. 

Artificial intelligence is one major reason for the boom. The technology powers all sorts of online experiences, from programs that predict your search query to apps that identify birds. But some experts are especially concerned about the more recent growth of generative AI. These tools, which can write emails, conjure images, and compose music based on a user’s prompt, require intense computing power—and vast quantities of electricity. Experts warn that AI’s insatiable appetite could derail efforts to ditch fossil fuels and confront climate change. “Decarbonization is at risk,” says University of Chicago computer scientist Andrew A. Chien. “Progress is at risk.”

“Decarbonization is at risk. Progress is at risk.”

Loudoun County epitomizes these tensions. Power demand grew in the county by 340 percent between 2018 and 2023. To feed data centers, the regional grid operator plans to keep running coal plants in West Virginia and Maryland that were previously scheduled for retirement. Dominion Energy is building a natural gas plant 100 miles to the south for the same purpose, despite local opposition. 

Generative AI requires tremendous energy because it churns through extraordinarily complex math. Every line of text from a chatbot requires billions of calculations, says Shaolei Ren, associate professor of electrical and computer engineering at the University of California, Riverside. Writing a single email with ChatGPT can use as much energy as driving a two-ton electric vehicle half a mile, Ren says. 

“AI is a fundamentally different computing paradigm,” Chien says. “It has this sort of limitless potential for growth that means you have to pay extra attention to it.” In 2022 data centers accounted for about 4.5 percent of U.S. energy demand. By 2030 that figure could reach 17 percent, due in large part to generative AI, per a Bloomberg analysis.

AI supporters say that, with the rise of carbon-free power sources, the technology doesn’t have to be an environmental problem. However, in the United States renewable energy production has grown by only about 1 percent per year since 2010, notes Chien. To offset worst-case energy demand forecasts for data centers alone, that rate would suddenly need to more than double. 

In response, companies are seeking out other sources of carbon-free energy. Constellation Energy and Microsoft are working to reopen the infamous Three Mile Island nuclear plant, with the aid of taxpayer dollars, solely to power data centers. Google plans to rely on small modular nuclear reactors, which have yet to be built. Meanwhile, Microsoft’s and Google’s embrace of data centers and AI has caused both companies’ carbon emissions to spike, according to their 2024 sustainability reports. 

Writing a single email with ChatGPT can use as much energy as driving a two-ton electric vehicle half a mile.

To put AI on a more responsible trajectory, regulators should require transparency and corporate accountability, says Priya Donti, an MIT computer scientist and cofounder of the nonprofit Climate Change AI. Loudoun County has moved in that direction, giving the county board more discretion over project approval. Congressional Democrats introduced legislation in 2024 to evaluate AI’s environmental impact, which also includes significant water consumption to keep servers from overheating. Running 10 ChatGPT queries requires up to 2 liters of water, Ren estimates.

Some proven solutions could also help, such as building data centers in places with plenty of water and carbon-free energy. Adding energy storage eases strain on the electric grid when the sun isn’t out and the wind isn’t blowing, and companies can go further by shifting their most energy-intensive tasks to times when there’s surplus. 

Yet for businesses to adopt sustainability measures, they’ll need external prodding: tax breaks for cutting emissions, more oversight on new development, or outright restrictions on when, where, and how data centers can be built. At this stage, all options should be on the table, Ren says. Otherwise, “I’m not hopeful that the tech industry will be able to solve this problem on its own.” 

This story originally ran in the Winter 2024 issue as “Does Not Compute.” To receive our print magazine, become a member by making a donation today.

 

Subscribe for the new deals