How AI Data Centers Are Spreading Air Pollution Across the US

AI's growing power demand is spreading air pollution far beyond data centers. See what new research finds on health impacts, where they're worst and why.

The rapid expansion of data centres to support artificial intelligence is worsening air quality and contributing to premature deaths in regions far from the facilities themselves, according to a new study by researchers at the Massachusetts Institute of Technology (MIT). The study estimates that electricity demand from US data centres in 2023 was associated with around 650 premature deaths from air pollution. While much of this demand was concentrated in states such as Virginia and Texas, Midwestern states saw some of the highest estimated health impacts. Researchers found that emissions from power plants supplying data centres can travel across state boundaries, exposing communities that host few data centres to pollution generated to meet distant electricity demand.

Data centre electricity demand drives pollution beyond local communities

The researchers examined data centre electricity demand across the United States in 2023, tracing how additional demand affected power generation, emissions, air quality, and public health. Their modelling framework combined an electricity-system model with an atmospheric chemical transport model to identify which power plants responded to data centre demand and where the resulting pollution travelled.

Data centres require electricity continuously to run computing operations, including artificial intelligence workloads. When demand rises, power plants must generate additional electricity, often using fossil fuels. The researchers found that generators responding to this additional demand were frequently natural gas-fired plants, although the mix varied by region and time.

The increased emissions included nitrogen oxides and sulphur dioxide, which contribute to the formation of ground-level ozone and fine particulate matter, or PM2.5. These pollutants can travel through the atmosphere, meaning the health effects of electricity consumption may be felt far from the facilities consuming the power.

Midwest bears a disproportionate health burden

The study estimated that data centre electricity demand contributed to 513 premature deaths associated with PM2.5 and another 137 linked to ground-level ozone in the continental United States in 2023. Together, the estimated 650 deaths represented around 5.3 per cent of the country’s total electricity-related mortality.

The geographical distribution of the estimated health impacts did not correspond closely with where data centres were concentrated. Northern Virginia and Texas accounted for much of the electricity demand, but some of the highest premature mortality rates were estimated in Iowa, Nevada, South Dakota and Kansas.

Researchers attributed this mismatch partly to the interconnected electricity grid. Additional demand in one region can trigger electricity generation elsewhere, including in areas with older and more polluting coal- and gas-fired power plants. Emissions from these facilities can then travel downwind, affecting neighbouring states.

Agricultural regions may face additional pollution impacts because ammonia released from fertilisers can react with pollutants from power plants to form fine particulate matter. The study estimated that health impacts extended to states including Nebraska, Missouri and Arkansas, despite these areas having relatively few data centres.

Rural communities face greater exposure risks

The researchers also found that rural populations were expected to face significantly higher mortality risks from data-centre electricity demand than urban populations. This suggests that communities located away from major technology hubs may still bear some of the environmental and health costs associated with the growing demand for digital infrastructure.

The study argues that assessments of data-centre impacts should extend beyond the immediate surroundings of facilities. Decisions about where to build new data centres and how to supply their electricity should account for regional power generation, atmospheric conditions and the distribution of pollution-related health burdens.

The researchers said future work could examine pollution from on-site power generation, which is increasingly being considered or used by data centres seeking dedicated electricity supplies. They also plan to investigate whether adjusting data-centre operations and electricity demand could reduce air pollution impacts. The findings add to concerns over the environmental footprint of AI infrastructure, suggesting that its effects are not limited to electricity consumption, water use or land requirements at the site of construction. The electricity needed to sustain these facilities can also contribute to pollution and health risks across wider regions.

References:

https://iopscience.iop.org/article/10.1088/1748-9326/aebbd4

https://phys.org/news/2026-11-centers-air-quality-distance-reveals.html

https://news.mit.edu/2026/study-reveals-data-centers-long-distance-damage-air-quality-1119

Banner image: Photo by Winston Chen on Unsplash 

Sections of this article may have been developed with the assistance of artificial intelligence tools to support research, drafting, and language refinement. All information has been reviewed, edited, and verified by the author/editor to ensure accuracy, context, and editorial integrity. The responsibility for the final content, interpretations, and conclusions rests solely with the publisher.

Vivek Saini
Vivek Saini
Articles: 360