...
By Jeff Young
Environment and Sustainability Editor
0
Share
Newsweek is a Trust Project member
See more of our trusted coverage when you search.Prefer Newsweek on Googleto see more of our trusted coverage when you search.
The energy required for artificial intelligence causes air pollution from fossil-fueled power sources, a new study has found, and the authors warn that as AI grows so will the health burden from air pollution unless data centers use cleaner power.
Researchers at the University of California, Riverside, and CalTech project that growing demand for power for the AI boom could result in as many as 1,300 premature deaths and billions of dollars in health costs in the U.S. each year by the end of this decade.
Despite growth in cleaner energy and progress in pollution controls, many parts of the country still rely heavily on coal and natural gas to generate electricity, and pollutants from power plants such as fine particulate matter and nitrogen oxides contribute to respiratory and cardiovascular diseases.
Read More on U.S.
The rapid growth of data centers to drive AI will be a major source of demand for electricity and thus a major contributor to air pollution, the researchers said.
...
"These are real impacts that we'll be seeing if we're not taking any actions right now," UC Riverside associate professor of electrical and computer engineering Shaolei Ren told Newsweek. "And those are preventable to a large extent."
Ren and his co-authors drew on industry data about energy consumption by data centers and used a tool developed by the U.S. Environmental Protection Agency to calculate the air pollution and health impacts from the electricity use.
However, many data centers also rely on backup generators separate from the electric grid. For example, when Elon Musk launched his xAI data center in Memphis, Tennessee, in July, it relied on a fleet of mobile generators, raising pollution concerns among residents there.
To get a fuller picture of data center power and pollution sources, the researchers looked to Virginia, home to the world's largest concentration of data centers. The state's environmental officials issue air quality permits for the generators data centers use, allowing Ren and his colleagues to estimate the pollution from those sources as well.
The combined health toll from air pollution associated with data centers as energy consumption rises was surprising, Ren said, and in some cases surpassed the pollution from other well-known problem areas.
"It turns out that AI data centers will be exceeding the entire California on-road emissions," Ren said. "So that's pretty surprising to me, and it suggests that we should shift our attention to AI."
The researchers estimated how much air pollution might arise from the computing work that goes into training some AI large language models. They found that training a model the size of the Llama 3.1, launched this summer by Facebook parent company Meta, could produce the same amount of pollution as 10,000 cars driving round-trip between Los Angeles and New York.
Ren said tech companies and data center developers should provide more information about air quality effects for a "more holistic, comprehensive evaluation of the impacts" on host communities.
The findings could quickly become more relevant to air quality concerns as data centers become a bigger part of U.S. electricity consumption. The U.S. Department of Energy and several regional electricity grid managers say they expect a surge in electricity demand due in part to data centers. The Electric Power Research Institute projects that by the end of the decade, data centers could consume about 9 percent of the electricity the country produces.
Newsweek hosted a panel discussion event during Climate Week NYC to explore the environmental impacts of AI and its soaring energy consumption. Panelists from AI leaders including Microsoft and Salesforce stressed their commitments to ambitious climate goals and investments to develop clean-energy sources for data centers.
Microsoft Vice President of Energy Bobby Hollis said AI's rapid growth means emissions are "growing slightly more quickly than we expected." Microsoft is investing in renewable energy development as well as nuclear power, including an agreement with Constellation Energy to restart a reactor at its Three Mile Island generating station in Pennsylvania.
Other tech companies are pouring money into clean energy as well. But even with that investment to produce cleaner electrons, fossil fuels are still playing a large role in powering data.
Meta announced last week that its largest new AI data center, a $10 billion facility planned for construction in Richland Parish, Louisiana, will be powered by natural gas. Regional power supplier Entergy plans to add advanced gas turbines to provide electricity for the data center with a lower emissions profile than that of a traditional natural gas power plant.
In an email to Newsweek, a Meta spokesperson said the company will also work with Entergy to bring at least 1,500 megawatts of new renewable energy to the regional grid.
"At Meta, we remain committed to maintaining net-zero greenhouse gas emissions for our global operations, building innovative and sustainable infrastructure, reporting transparently on our sustainability goal progress, and supporting the communities where we operate," spokesperson Ryan Daniels wrote.
Recent statements by Exxon Mobil and Chevron indicate the oil and gas giants also see an opportunity to benefit from the AI boom by providing gas to power data centers.
Chevron New Energies President Jeff Gustavson said at a Reuters conference Wednesday that the company has been working with tech companies on projects that would capture carbon emissions from burning natural gas, Reuters reported. The New York Times reported this week on Exxon Mobil's similar move into the data power space, "designing a massive natural-gas fueled plant meant to directly supply electricity to data centers," according to the Times.
The AI air pollution study published this week builds on Ren's earlier research into AI's environmental impact. As Newsweek reported in March, Ren did groundbreaking work on the astonishing water consumption by data centers as they cool computer servers.
Several major tech companies and data center developers stress their efforts to address greenhouse gas emissions and water usage, and annual sustainability reports focus on work to improve efficiency and develop cleaner operations. Ren said he was motivated to do the new work because toxic air emissions were not getting the same attention.
"We also have something that is already affecting people right now, and that is air pollution," he said.
The new study appears in arXiv, a curated research-sharing platform maintained by Cornell University. According to the site, arXiv moderators "check for scholarly value" in submitted work, but studies shared there are not peer-reviewed.
Other researchers investigating AI's energy use cautioned that there is a high degree of uncertainty in projections about AI's potential growth and that AI model development and energy efficiency in computing could change dramatically. Such variables make projections about AI's energy consumption and pollution profile tricky.
But even with those caveats, some scientists applauded the study as another step toward better understanding AI's potential effects.
"AI's environmental impacts are significant—from the water and energy required to the greenhouse gases emitted by the different stages of the AI lifecycle," Sasha Luccioni, climate lead at the tech research company Hugging Face, told Newsweek via email. "It's really important to get more transparency and accountability when it comes to these impacts and how they're liable to evolve in the future."
View 0 Comments
Submit Correction
View Editorial & AI Guidelines