A rusted rooftop weather vane points toward homes beneath a power plant plume, beside a cutout of a server rack.
Image by CryptoSlate

AI data center boom linked to 650 premature US deaths as pollution spreads across states

MIT researchers found that emissions from power plants supplying data centers can harm distant communities, while Harvard estimates 25.7 million tons of carbon emissions.

Quick Take

  1. MIT researchers linked air pollution from power plants supplying US data centers to an estimated 650 premature deaths in 2023.
  2. Pollution traveled across state lines, leaving communities with few data centers facing some of the highest estimated death rates.
  3. As AI infrastructure expands, researchers warn that new facilities could impose health costs on distant communities unless power sources are considered.

Air pollution from power plants supplying US data centers contributed to an estimated 650 premature deaths in 2023, researchers found.

An MIT-led study published Nov. 19 found that pollution associated with data center electricity demand traveled across state boundaries, exposing communities with relatively few facilities to elevated health risks.

The findings follow separate Harvard University research estimating that major US data centers generated 25.7 million metric tons of carbon dioxide emissions over a 12-month period, highlighting the environmental costs accompanying the artificial intelligence industry's expansion.

Data center pollution reaches communities hundreds of miles away

The MIT researchers examined how data centers' electricity demand influenced power generation and air pollution across the contiguous United States in 2023.

Their findings, published in Environmental Research Letters, attributed 513 premature deaths to fine particulate pollution and another 137 to ground-level ozone.

Together, those deaths represented approximately 5.3% of the country's estimated mortality linked to electricity-related air pollution.

How data center power use affects downwind air

The 2023 model followed electricity demand across the grid and tracked how the resulting emissions moved through the air.

Modeled premature deaths by pollutant

Contiguous US, 2023. The study estimated 513 premature deaths from fine particles and 137 from ground-level ozone.

The place that draws the power and the place exposed to its pollution can be in different states.

The diagram illustrates the study’s mechanism without mapping any single plant or route. The chart shows modeled counts for 2023 grid electricity and excludes onsite backup generation.

Virginia and Texas accounted for the largest data center electricity demand, consuming 33.9 and 21.8 terawatt-hours, respectively. However, Iowa, Nevada, South Dakota, and Kansas recorded some of the highest estimated death rates per resident, despite several having relatively few data centers.

The geographical imbalance reflects how America's interconnected electricity grid distributes the environmental consequences of growing power demand.

For example, electricity consumption by Virginia data centers can increase generation at coal and natural gas plants in the Ohio Valley. Emissions from those facilities subsequently travel through the atmosphere, where they can combine with ammonia from agricultural activities in the Midwest to form harmful fine particles.

These pollutants are associated with heart attacks, strokes, and respiratory illnesses.

The researchers also found that rural populations faced significantly higher estimated mortality risks than urban communities, extending the health consequences beyond the regions hosting major computing facilities.

The analysis tracked how power plants responded to data center electricity demand and used atmospheric models to estimate where the resulting pollution accumulated.

Harvard research exposes broader environmental costs of AI expansion

A separate Harvard-led study published Nov. 16 in Nature Sustainability examined electricity consumption and carbon emissions across 403 major US data centers between May 2024 and April 2025.

The researchers estimated that the facilities consumed approximately 82 terawatt-hours of electricity, equivalent to 1.8% of total US electricity consumption.

That demand was associated with approximately 25.7 million metric tons of carbon dioxide emissions, with fossil fuels accounting for roughly 54% of the electricity generation attributed to the facilities.

Depending on assumptions about electricity demand, estimated emissions ranged from 21 million to 31 million metric tons.

The study also found that the facilities' average carbon intensity was approximately 314 grams of carbon dioxide per kilowatt-hour, compared with the national electricity generation average of 348 grams.

The Harvard findings add to concerns about the environmental costs of hyperscale computing infrastructure as technology companies expand capacity to meet growing demand for AI services.

Related Reading

America's $1 trillion AI data center boom just suffered its first slowdown in over a year

While Harvard examined greenhouse gas emissions, MIT traced how electricity generation contributes to air pollution and premature mortality.

The findings intensify scrutiny of where new data centers are built and how their electricity is generated.

MIT researchers urged planners to consider air-quality consequences across affected regions, including communities far from the facilities, when deciding where to locate new computing infrastructure.