This story was originally published by Mother Jones and is reproduced here as part of the Climate Desk collaboration.
While tech oligarchs grandstand about the existential danger AI superintelligence poses to humanity, former Environmental Protection Agency officials say we should be worried about a more immediate threat to our health: the data centers that power those AI models.
The former EPA researchers, now working with the nonprofit Environmental Protection Network, are warning in a new report that since January 2025, Donald Trump’s EPA has embarked on a systematic weakening of pollution safeguards, permitting processes, inspection norms, and other guardrails — at least 30 different actions — that could increase the health risks associated with the data center boom.
There are currently about 3,000 data centers in the United States, and more than 1,500 new centers are in the planning stages. But as developers build out this vast, energy-hungry infrastructure, the country is losing the government regulatory safeguards that are needed to ensure that expansion is done safely, the ex–EPA officials say.
Dan Costa, who spent 35 years at the EPA studying air pollution, noted that for decades the environmental agency, which was established in 1970, had a large and committed cadre of researchers who investigated emerging sources of pollution and new environmental threats.
“Over the decades, toxicologists, chemists, atmospheric scientists, modelers, and engineers worked in concert with academic colleagues on behalf of the American people — not on behalf of industry,” Costa said in a press briefing last week. “We wanted to carry out EPA’s mission of protecting people’s health from industrial air pollution.”
Now, though, that workforce has been gutted. The Trump administration has dismantled EPA’s main research office, and the agency’s priorities have shifted away from strict enforcement of pollution standards. “Transforming the nation’s energy system, while diminishing its scientific capacity to understand what the impacts are going to be, makes no sense,” Costa said.
But that appears to be the direction the country is heading in. President Donald Trump is “positioning America as the global leader in AI,” EPA Administrator Lee Zeldin wrote in a memo last year. The agency is “committed to bolstering” those efforts, Zeldin said, by “cutting heavy-handed regulations to ensure America can build and power the data centers essential to Making America the AI Capital of the World.”
Since Trump returned to the White House, data center developers have seized the opportunity to expand at a rapid pace, helped by the administration’s moves. In some cases, that deregulation has involved keeping old coal-fired plants online past their planned decommissioning times in order to power data centers; other times, it has meant easing regulations on the use of onsite fossil fuel-powered turbines to power hyperscale data centers like xAI’s Colossus.
The agency has proposed a rule change that would let developers begin constructing data centers and their associated power plants before they get air pollution permits. And it has directed the agency’s enforcement staff to avoid any action that could “unduly burden or significantly disrupt” energy production or power generation.
According to one study the former EPA researchers cited in their new report, pollution associated with the AI boom may lead to 1,300 additional deaths per year.
Physician and former EPA administrative official Lynn Goldman said some health impacts of increased data center pollution might be felt quickly — for example, heart attacks or premature births associated with particulate exposure. Others, such as cancer, might take years to emerge. “The AI industry’s race for advantage should not be a matter of life or death,” Goldman said at the press conference. “Families shouldn’t have to pay for the AI boom with their health.”
Facts Only
* There are approximately 3,000 data centers in the United States.
* More than 1,500 new data centers are currently in the planning stages.
* Donald Trump is the President of the United States.
* Lee Zeldin is the EPA Administrator.
* Since January 2025, the EPA has taken at least 30 actions to weaken pollution safeguards and permitting processes.
* The EPA proposed a rule change allowing data center construction to begin before air pollution permits are issued.
* The administration has allowed some coal-fired plants to remain online past planned decommissioning dates to power data centers.
* Regulations regarding onsite fossil fuel-powered turbines for hyperscale data centers, including xAI’s Colossus, have been eased.
* The EPA's main research office has been dismantled.
* One cited study estimates pollution from the AI boom may cause 1,300 additional deaths per year.
Executive Summary
The rapid expansion of AI infrastructure in the United States is creating a tension between economic competitiveness and public health protections. Under the Trump administration, the EPA is prioritizing the goal of making America the "AI Capital of the World" by reducing regulatory burdens on data center developers. This strategy includes accelerating construction timelines by delaying permit requirements and utilizing older fossil-fuel energy sources to meet massive power demands.
Former EPA officials and researchers warn that this deregulation—specifically the dismantling of scientific research offices and the weakening of air pollution safeguards—poses immediate health risks. They cite potential increases in heart attacks, premature births, and long-term cancer risks, with some estimates suggesting up to 1,300 additional deaths annually. While the administration views these moves as essential for global leadership in AI, critics argue that the speed of industrial growth is outpacing the government's capacity to monitor and mitigate its environmental impact.
Full Take
The strongest version of this narrative is a classic conflict between rapid industrialization and the "precautionary principle." It posits that the pursuit of geopolitical AI dominance is being subsidized by the physical health of local populations, shifting the cost of innovation from the developer to the public.
The narrative utilizes a specific load-bearing pattern: the juxtaposition of "existential AI danger" (the abstract, futuristic fear) against "immediate health threats" (the concrete, present danger). This frames the tech industry's own warnings about AI superintelligence as a diversionary tactic to mask tangible environmental degradation. By contrasting the "oligarchs" with "families," the argument moves from a policy debate to a moral one.
Patterns detected: ARC-0024 Ambiguity (the term "safeguards" is used broadly to cover both scientific research and specific permitting rules).
The root cause is the recurring historical pattern of "Regulatory Capture" or "Accelerationism," where the perceived urgency of a new technological frontier is used to justify the suspension of established safety norms. This echoes previous eras of rapid industrial expansion where environmental costs were externalized to ensure market dominance. The second-order consequence is the erosion of institutional memory; once the scientific cadre of an agency is dismantled, the ability to detect new threats is lost, regardless of who holds power.
Bridge Questions:
1. What specific metrics would be required to prove a direct causal link between data center deregulation and the cited mortality rates?
2. How do the projected economic benefits of AI leadership weigh against the quantified health costs in a formal cost-benefit analysis?
3. Are there alternative energy scaling models that could achieve the same AI capacity without relying on decommissioned coal plants?
Counterstrike Scan: A coordinated influence campaign would use "Fear Appeal" by amplifying the 1,300 deaths figure without providing the underlying study's methodology to trigger public panic. The actual content remains focused on the policy shift and the experts' warnings, remaining within the bounds of standard journalistic reporting.