AI data center health risks are the focus of a new report from former Environmental Protection Agency officials, who identified at least 30 federal actions they say weaken protections as computing infrastructure expands. Seventeen of those actions explicitly cite or target artificial intelligence or data centers.

 

The September 10 report separates three policy pressure points:

  • Weaker or reconsidered pollution safeguards.
  • Faster permitting with less public review.
  • Reduced scientific, monitoring and enforcement capacity.

 

AI Data Center Health Risks Span Local and Regional Sources

The Environmental Protection Network report argues that pollution can originate both at data-center campuses and far beyond them. On-site diesel generators and gas turbines affect nearby communities, while distant power plants supply the much larger electricity loads created by AI computing.

 

EPN is an independent group comprising more than 800 former EPA scientists, engineers, toxicologists, policy analysts and other professionals. Its report reviews federal actions taken since January 2025 and assesses how they may change pollution safeguards, project approvals and the government’s ability to measure and enforce environmental requirements.

 

The report cites a separate study by researchers from UC Riverside, Caltech and Rochester Institute of Technology. That model estimates data-center-related air pollution could create annual public-health costs of $11.7 billion to $20.9 billion by 2028.

 

Under the study’s high-growth scenario, researchers estimated about 600,000 asthma symptom cases and 1,300 premature deaths. More than 90 percent of the modeled health effects came from power plants serving data centers, showing why the relevant footprint extends beyond the boundaries of individual facilities.

 

Those figures are projections, not a measured toll from the 30 actions. EPN explicitly says the underlying study does not calculate the additional effect of the policies in its new report. The organization’s central criticism is that no public federal analysis currently fills that gap.

 

The 30 Federal Actions Cover Permits, Power and Enforcement

EPN divides the actions into three broad groups. The first includes decisions affecting limits on soot, mercury, toxic air pollution, interstate smog, oil and gas emissions, coal wastewater and coal ash, along with steps that keep some fossil-fueled generating capacity operating longer.

 

The second group concerns permitting and public participation. The report argues that projects can move ahead before major pollution questions are resolved, while residents receive less information or fewer opportunities to influence decisions before developers make costly commitments.

 

The third group addresses government capacity. Staff and funding reductions can affect emissions measurement, risk assessment, facility inspections and enforcement. The effect is difficult to quantify because a legal standard has limited value when agencies lack the people or data needed to check compliance.

 

One official example is EPA’s July guidance for “islanded” power generation. The agency concluded that facilities not connected to the public grid do not fall under the Clean Air Act’s Acid Rain Program because they do not sell electricity or report as generating units to the Department of Energy.

 

EPA Defends Its AI Data Center Permitting Approach

EPA says the islanded-power interpretation gives data-center developers more flexibility while reducing burdens on the electric grid. The agency also says a facility could become subject to the Acid Rain Program if it later connects to the grid.

 

In response to EPN, an EPA spokesperson told The Verge that the agency is returning regulations to what it considers the best reading of the Clean Air Act. EPA says it remains committed to protecting health and the environment while making the United States the leading location for AI development.

 

The dispute therefore concerns both the legal reach of existing rules and the practical consequences of faster infrastructure approvals. Developers want predictable paths to power new campuses, while critics argue that exemptions and reduced oversight can shift health costs from technology companies to surrounding communities and power-plant regions.

 

For AI companies, the issue is becoming a supply-chain and siting risk. Electricity contracts, backup generation, local air permits and grid connections can all affect project schedules. Environmental challenges may also create litigation, disclosure and community-relations costs even when a federal agency has streamlined part of the review.

 

EPN Proposes a Data Center Health Protection Pledge

EPN is calling for a Data Center Health Protection Pledge modeled on the White House’s Ratepayer Protection Pledge, which asks AI companies and power providers to prevent households from bearing higher electricity costs. The new proposal would extend that principle to pollution and public health.

 

The group wants agencies to restore and enforce pollution safeguards, properly permit and monitor on-site power, preserve meaningful public review and rebuild scientific and enforcement capacity. It also wants public reporting on how the data-center buildout changes emissions, who is exposed and what those exposures mean for health.

 

A pledge would not by itself create binding emission limits. Its value would depend on measurable commitments, public data and enforcement mechanisms. The report’s most immediate challenge to EPA is therefore methodological: quantify how rapid AI infrastructure growth and current policy choices jointly affect air quality.

 

The next evidence to watch will be facility-level emissions data, regional power-generation changes and court decisions testing the government’s interpretations of environmental law. Those records can show whether faster approvals coexist with stable pollution levels or whether the health burden projected by researchers begins to materialize.

 

Background Reading