Trump Exempts Data Centers From Pollution Rules, Fueling AI Expansion

A coalition of former EPA officials warned this week that the Trump administration's campaign to strip environmental rules in service of AI infrastructure expansion could cost Americans up to 1,300 premature lives and more than $20 billion in public health costs by 2028, a reckoning that institutional capital has not yet priced into AI infrastructure positions.

The warning arrived September 12, 2026, in a report from the Environmental Protection Network, an independent organization of EPA alumni. The group identified 30 distinct federal actions taken since January 2025 that it says have worsened the health risks tied to data center construction and operation. Of those 30 policy changes, 17 specifically name AI or target data centers as the rationale for regulatory relief. The underlying health and mortality estimates come from a study conducted jointly by UC Riverside, Caltech, and the Rochester Institute of Technology.

"You would think that an industry buildout of this magnitude, billed as a revolution, would mean more vigilance from EPA," Lynn Goldman, a pediatrician and former EPA assistant administrator, said at the briefing. "Instead, EPA is doing just the opposite, thinking first about industry's needs and weakening or reconsidering limits on toxic chemicals and harmful pollutants."

For anyone without a stake in data center real estate or AI infrastructure bonds, this matters because the deregulation architecture being constructed in Washington is quietly transferring costs from technology companies onto the public balance sheet. The $20 billion figure is not a speculative ceiling. The EPN report states explicitly that the true toll could exceed that number once the 30 identified policy rollbacks are factored into the UC Riverside baseline. That is a liability with no current owner in any AI infrastructure prospectus.

The 30 Rollbacks: A Systematic Dismantling, Not a Clean Slate

The scale of the deregulation effort is deliberate and document-backed. EPA Administrator Lee Zeldin, appointed at the start of Trump's term, stated his objective plainly: making America "the AI capital of the world" through deregulation. Since that declaration, Zeldin and the White House have rolled back dozens of rules while simultaneously cutting EPA staff and funding.

Trump's "AI Action Plan," released in July 2025, recommended streamlining or reducing regulations under the Clean Air Act, the Clean Water Act, and Superfund law. The explicit purpose was to expedite permitting for data center projects and chip factories. Tyson Slocum, energy program director at the consumer advocacy group Public Citizen, told The Verge at the time that the AI Action Plan was "using unprecedented emergency powers to grant massive new exemptions for data centers and specifically fossil fuel infrastructure."

The policy groundwork has translated into physical infrastructure. The administration has laid the foundation for new gas projects to power energy-hungry data centers, ranging from turbines installed on-site to dedicated power plants built specifically to serve technology companies. At the same time, surging demand for advanced chips is reviving production of per- and polyfluoroalkyl substances, the so-called forever chemicals that accumulate in human bodies and resist environmental breakdown.

The EPA's response to the EPN report is on record. Spokesperson Cora Mandy told The Verge in an email: "In each action EPA has taken, the agency has returned regulations to the best reading of the Clean Air Act after years of overreach by previous administrations." That framing, returning to a best reading, is the administration's legal defense. It is not a denial that pollution increases.

The Geography of Harm: Pollution Does Not Stop at the Data Center Fence

One analytical framing that the source report makes explicit, and that equity markets have not absorbed, is the diffuse geography of AI-related pollution. Goldman stated at the briefing: "You do not need to be near a data center to be affected, because the pollution doesn't just come from the data centers themselves."

This is not a nuance. It is the central investment risk. The AI infrastructure buildout is frequently modeled as a local permitting problem, a question of whether a specific campus in Virginia or Texas can clear environmental review. The EPN analysis repositions it as a systemic atmospheric and chemical contamination issue that spreads through power grid emissions, forever chemical runoff, and particulate matter carried by wind far beyond any single facility.

The implication for institutional capital: the liability exposure for data center operators is not geographically bounded. Class action and regulatory risk, if the political cycle turns, attaches to the entire grid draw and supply chain, not just the physical footprint of a given campus.

Last year, hundreds of current EPA employees signed a public letter to Zeldin stating that the agency under his leadership would "not protect communities from hazardous chemicals and unsafe drinking water, but instead will increase risks to public health and safety." That letter was a contemporaneous internal signal, not retrospective criticism. The institutional community largely ignored it.

What the Deregulation Subsidy Is Really Worth, and Who Collects It

Follow the money. The regulatory rollbacks are, in economic terms, a subsidy transfer. Environmental compliance costs for fossil fuel power generation, chemical manufacturing, and large industrial facilities are real and quantifiable. When the government exempts those facilities from standards, the cost does not disappear. It migrates to public health systems, to individuals with no market recourse, and ultimately to government budgets through healthcare expenditures.

The $20 billion public health cost estimate through 2028 represents the floor of that transfer, per the UC Riverside, Caltech, and Rochester Institute of Technology study cited by EPN. The beneficiaries of the transferred cost are technology companies building data centers and the fossil fuel operators powering them. Neither currently carries that liability on a balance sheet.

Marc Boom, EPN's senior director of public affairs, stated at the briefing: "A more polluting AI buildout is not inevitable. It depends on the choices government makes now. EPA should be measuring how this buildout is changing pollution, who is being exposed, what it means for their health, and making all of that information public."

The absence of that measurement regime is itself an investment risk. Without federal data collection, the liability remains diffuse and unattributed. The moment measurement resumes, attribution follows.

The Governance Gap That AI Infrastructure Stocks Are Not Pricing

The AMD rack-scale expansion and ServiceNow's AI Control Tower rollout, both announced in the week of September 11, 2026, illustrate the pace at which enterprise AI infrastructure investment continues to accelerate. Neither announcement addresses the regulatory and environmental liability framework within which that infrastructure operates. That is the governance gap.

Institutional investors running ESG mandates face a specific problem. The deregulatory actions documented by EPN create an environment where the traditional metrics used to assess environmental compliance, EPA enforcement actions, permit conditions, air quality monitoring data, are being systematically dismantled as measurement tools. An investor trying to conduct environmental due diligence on a data center operator today has less publicly available federal data than at any point in the past decade, by design.

The Plocamium View

The market is treating the EPA deregulation story as a political narrative with no financial content. That is wrong, and the error is measurable.

Here is the thesis: the Trump administration's rollback of 30 environmental rules since January 2025 is creating a contingent liability that will crystallize in the 2028 to 2032 window. The mechanism is straightforward. A future administration, state attorneys general, or class action plaintiffs armed with the UC Riverside and Caltech research will have a documented body of evidence linking specific policy decisions to specific health outcomes. The $20 billion public health cost figure is the academic foundation for that litigation. The 30 named policy actions are the jurisdictional roadmap.

Data center operators, hyperscale cloud providers, and the private equity funds that own the physical infrastructure assets beneath them are accumulating unbooked environmental liability at the precise moment valuations are highest. The AI infrastructure buildout has driven data center asset pricing to levels that assume minimal regulatory friction indefinitely. That assumption is the alpha opportunity.

Plocamium's second-order read: the geography of harm matters for asset selection. Facilities dependent on dedicated fossil-fuel power plants, the specific configuration that Trump's policy groundwork actively enables, carry higher tail liability than facilities with genuine renewable power purchase agreements. The former are cheaper to build today because the environmental cost is externalized. They will be more expensive to own in five years when that cost is internalized through litigation, carbon pricing, or reinstituted EPA enforcement.

The "Data Center Health Protection Pledge" that EPN is urging the Trump administration to adopt will not happen in this political cycle. But the organizations and legal frameworks it would institutionalize already exist. The EPN itself, alongside the public letter from hundreds of active EPA employees, represents institutional knowledge and institutional intent that does not dissolve with a change in administration. It accelerates.

For infrastructure allocators, the actionable question is not whether deregulation is good or bad for returns in 2026. It is whether the current vintage of data center assets is being underwritten with a discount rate that reflects a litigation tail beginning in 2028. Our read is that it is not.

The Bottom Line

The EPN report, published September 12, 2026, is the first systematic, source-documented accounting of how AI infrastructure deregulation translates into human health costs. The number, more than $20 billion and up to 1,300 premature deaths by 2028, is not a protest figure. It is a peer-reviewed academic estimate that will form the foundation of future regulatory and legal action.

Institutional capital positioning in AI infrastructure must now include a litigation discount. The facilities most exposed are those powered by new gas infrastructure, operating in jurisdictions with weakened air quality enforcement, and owned by entities that cannot demonstrate good-faith environmental compliance independent of federal mandates. The deregulation subsidy is real and it is being collected today. The bill will arrive on a future watch. The only question is who holds the asset when it does.