Study links data-center power demand to an estimated 650 premature deaths in the US
The highest estimated per-person impacts were not concentrated in the biggest data-center hubs. An hourly grid model helps explain the mismatch.
Loading page…
The highest estimated per-person impacts were not concentrated in the biggest data-center hubs. An hourly grid model helps explain the mismatch.
Listen to this story
By tracing hourly changes in output across more than 10,000 US generators, an MIT-led analysis links data-center electricity use in 2023 to an estimated 650 premature deaths—about 5.3% of electricity-related mortality. The modeled burden was not confined to server hubs: several Midwestern states had higher per-person impacts than Virginia or Texas, and rural populations faced greater estimated risk. The results make power sourcing and grid geography relevant to siting decisions, but do not count identified deaths; on-site generation and flexible demand remain for future study.
The model attributed 513 estimated deaths to fine-particle pollution and 137 to smog; these are epidemiological estimates, not individually verified fatalities.
Iowa, Nevada, South Dakota and Kansas had among the highest per-person impacts, while rates in Virginia and Texas were roughly half those in most Midwestern states.
Added Virginia-area demand can call on dirtier Ohio Valley generators, while agricultural ammonia helps form fine particles downwind in Nebraska and Iowa.
Living far from a data center does not necessarily mean escaping its air pollution. An MIT-led study published November 19 in Environmental Research Letters estimates that electricity demand from US data centers contributed to about 650 premature deaths in 2023, with some of the highest per-person impacts far from major computing hubs.
Many earlier studies estimated data centers’ effects using average grid emissions. This team instead modeled which generators increase output when data centers need electricity, tracking that response hour by hour. Like the earlier studies, it assumed the facilities draw power around the clock.
The researchers built an electricity-system model that simulates each generator’s hourly production and power transfers between regions. Gathering and checking the inputs meant dealing with more than 10,000 generators across the US grid, according to MIT graduate student and study coauthor Yuang (Albert) Chen.
That produced generator-level estimates of nitrogen oxides and sulfur dioxide emissions. Nitrogen oxides help form ground-level ozone, or smog; both pollutants contribute to fine particulate matter, known as PM2.5. A second model then simulated how pollution forms and travels through the atmosphere.
Finally, the team used evidence from epidemiological studies to translate the additional pollution into estimated premature mortality. These are modeled health effects, not a count of individually identified deaths. The researchers estimate the total represented about 5.3% of US electricity-related mortality in 2023.
The model attributed 513 premature deaths to fine particulate pollution linked to data-center electricity demand in 2023.
The model attributed another 137 premature deaths to smog linked to that electricity demand.
Much of the electricity demand was concentrated in northern Virginia and Texas. Yet Iowa, Nevada, South Dakota and Kansas had among the highest estimated premature mortality rates per person. Rates in Virginia and Texas were only about half those in most Midwestern states.
The interconnected grid explains part of that mismatch. Chen says generators responding to added demand in Virginia are likely to be in the Ohio Valley. That region has a higher proportion of older coal- and natural-gas-fired plants with greater emissions. The generator meeting a facility’s extra demand may therefore be far from its servers.
Atmospheric conditions add another layer. Pollutants can travel into downwind states, while ammonia from agricultural fertilizer helps fine particles form in places such as Nebraska and Iowa. The estimated health effects also reach Missouri and Arkansas, illustrating how the burden extends into states with relatively few data centers.
The researchers also found that rural populations face significantly higher estimated mortality risk from data-center electricity demand than urban populations. Fine particulate pollution can cause heart attacks, strokes and lung disease—the health consequences behind a geographic pattern that would be missed by looking only at a facility’s immediate neighbors.
Chen argues that data-center siting and operating decisions should account for air quality alongside land and energy prices. The study offers a baseline for considering both where facilities should be built and how their electricity should be supplied, he told MIT News.
The researchers next want to examine on-site generation, where power plants supply data centers directly. They could also study whether adjusting computing facilities’ electricity demand over time changes air quality. Those questions remain future work; the current findings concern the health effects traced through the interconnected grid.
Loading discussion...
Join the conversation
Explain how far a community’s influence should reach when pollution crosses borders.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.