The rapid growth of data centers built for artificial intelligence harms air quality and health in regions far from the buildings themselves, according to a new study by MIT researchers. The team estimates that extra air pollution from the electricity these facilities demanded in 2023 led to about 650 premature deaths across the United States.
The research appears in Environmental Research Letters. Some Midwestern states had among the highest estimated death rates per capita, even though most of the data center electricity demand was concentrated in Virginia and Texas, MIT News reported.
The model estimates that data center emissions caused about 650 premature deaths in 2023: 513 from fine particulate matter (PM2.5) and 137 from smog. That is about 5.3% of the nation’s total electricity-related mortality, according to the team.
Per-person death rates were highest in Iowa, Nevada, South Dakota and Kansas. Rates in Virginia and Texas were only about half as high as in most Midwestern states.
The U.S. power grid is interconnected, so a new source of demand in one place can make power plants in another place produce more. Doctoral student Yuang (Albert) Chen, a co-author, said the generators that respond to extra demand are often natural gas plants, though this varies by place and hour.
The researchers say that power plants answering demand in Virginia are likely in the Ohio Valley, where older plants that burn coal and natural gas make up a larger share of the mix. Noelle Selin, an MIT professor and the study’s senior author, said that plugging in a major new source of demand may force operators to fire up a very dirty plant.
Wind then carries many of the pollutants to downwind states. In agricultural states such as Nebraska and Iowa, the researchers also point to local conditions that affect how pollution forms, though MIT’s account of that point is cut short.
Many earlier studies relied on average grid emissions. This team, like those studies, assumed data centers draw power around the clock for computing. It then went a step further and identified which generators respond to that added demand hour by hour, so the emissions and health effects could be traced to specific places.
The researchers built a model of the electricity system that simulates how much each generator produces every hour and how power moves between regions. It yielded emissions of nitrogen oxides and sulfur dioxide for each generator. Chen said there are more than 10,000 generators on the U.S. grid, and the team had to check that the data for each was valid.
Those figures went into a chemical transport model that simulates how pollutants form and travel through the atmosphere. The team then converted the added pollution into premature deaths using findings from epidemiological research.
Fine particulate matter can cause heart attacks, stroke and lung disease, and nitrogen oxides contribute to ground-level ozone, or smog. The study suggests that the health costs of data centers are spread across a wide region and do not fall only on nearby neighbors.
“When people think about the negative impacts of data centers, they usually focus on the neighbors,” Selin said, adding that residents without a data center nearby should “think again.” The researchers say the results show why the location and operation of these facilities matter for public health on a broader scale.
The Lawrence Berkeley National Laboratory estimated that data centers accounted for about 4% of total U.S. energy demand in 2023 and about 4.4% of the nation’s overall emissions. That share has likely grown since, because new data centers continue to be built at a rapid pace, according to MIT News.
The study’s authors include researchers from MIT, the University of California at Irvine and Yale University. The 650 figure is a model estimate, not a count of individual deaths.
Around Prague, based on reporting by MIT News