Roundtable #37: Ground Truth: AI Data Centers and the Environmental Cost of Federal Inaction
Section I: Breaking Ground—Cumulative Harm Under Federal Environmental Law
In 1979, a study of solid waste disposal in Houston found that all five city-owned landfills and six of the city’s eight incinerators were located within Black neighborhoods. Three years later, North Carolina placed 40,000 cubic yards of soil contaminated by illegal polychlorinated biphenyl dumping in Afton, a rural Black community, despite years of legal challenges and six weeks of protests. Although the landfill itself was legally permitted under the Toxic Substances Control Act, the decision to site it in a majority-Black community with a twenty-five percent poverty rate became a defining example of how environmental law can authorize the concentration of harm in vulnerable communities. These incidents reflect a broader pattern: the concentration of polluting infrastructure in minority communities shaped by decades of residential segregation and government disinvestment. Environmental-justice organizing pushed the federal government to investigate racial disparities in hazardous-waste siting and, eventually, to incorporate environmental justice into federal decision-making.
By issuing Executive Order 12898 in 1994, President Bill Clinton directed federal agencies to identify and address “disproportionately high and adverse” environmental and health effects on minority and low-income populations. Although individuals could not sue the government solely for violating the Order, EO 12898 supplied a government-wide framework for incorporating environmental justice into agencies’ exercise of existing statutory authority. Residents seeking judicial relief still had to connect environmental-justice harm to a separately enforceable legal duty, most often under the National Environmental Policy Act (NEPA), which requires environmental review and disclosure; Title VI of the Civil Rights Act, which addresses discrimination in federally funded programs; or the Clean Air Act (CAA), which provides permitting and enforcement tools. Within those statutory processes, the Order encouraged agencies to identify overburdened communities, consider existing and proposed pollution together, and document disproportionate effects. President Donald Trump revoked EO 12898 on January 21, 2025, through Executive Order 14173, leaving the underlying environmental statutes intact while removing the government-wide directive. The revocation coincided with the federal government’s acceleration of artificial intelligence (AI) infrastructure development.
Data centers may satisfy individual environmental requirements while adding emissions, noise, water depletion, and public-health risks to already overburdened communities. EO 12898 did not give communities the power to block such projects, but it required agencies to identify who would bear the burden and to account for that distribution in their decision-making. Without it, the question of cumulative impact no longer has a guaranteed place in the federal process. The post-EO 12898 landscape leaves communities with the right to be heard without possessing a meaningful right to refuse. Federal law contains no general rule determining when cumulative environmental burden becomes legally unacceptable; harm generally becomes actionable only when residents can connect it to a separate statutory violation. The resulting gap allows an AI infrastructure project to be lawful piece by piece, even when its cumulative effect imposes a serious additional harm on the surrounding community.
xAI’s $20 billion Southaven data center, described by Governor Tate Reeves as “the largest economic development project in Mississippi’s history,” highlights this gap. Mississippi approved the project and granted economic-development incentives—including tax exemptions for computing equipment and software—while xAI, through a subsidiary called MZX Tech, mounted natural-gas turbines near homes, schools, and churches in North Mississippi and the broader Memphis area. The Mississippi Department of Environmental Quality (MDEQ) determined that because the turbines were “portable” and intended to remain on site for fewer than twelve months, they qualified as “nonroad engines” exempt from Clean Air Act permitting requirements. In April 2026, the NAACP filed a CAA citizen suit authorized by Section 304, 42 U.S.C. § 7604, alleging xAI contracted and operated twenty-seven natural-gas turbines without any applicable construction or operating permit. The core dispute is whether trailer-mounted turbines operating continuously at a single location qualify as portable nonroad engines exempt from permitting, or whether they meet the Clean Air Act’s definition of a stationary source as equipment that is “not self-propelled or intended to be propelled.” The complaint invokes three distinct permitting regimes. First, it alleges that xAI constructed a “major stationary source” without undergoing preconstruction review required by the Prevention of Significant Deterioration (PSD) program and Mississippi’s corresponding State Implementation Plan (SIP). Second, it argues that the facility’s potential emissions of nitrogen oxide exceed major-source thresholds and therefore require a comprehensive operating permit under Title V of the CAA. Third, it alleges violations of federal Hazardous Air Pollutant (HAP) standards, citing unpermitted emissions of carcinogens such as formaldehyde.
Even complete compliance with those permitting standards would not resolve the cumulative-harm question. The turbines have been alleged to emit more than 1,700 tons of nitrogen oxide, 500 tons of carbon monoxide, and nineteen tons of formaldehyde annually. Regulators may determine whether those emissions cross statutory thresholds and impose controls, but a valid permit establishes compliance with the conditions governing that source, not that adding its emissions to pollution from nearby highways, warehouses, or industrial sources is acceptable for the surrounding community. Federal law contains no general cumulative-burden threshold requiring regulators to reject a project because its additional pollution would impose an unacceptable combined burden on an already overburdened community.
NEPA offers an incomplete safeguard because it may not apply at all, and even when it does, it regulates decision-making rather than outcomes. Section 102(2)(C) of NEPA requires federal agencies to evaluate the environmental effects of major federal actions, but substantial consequences alone do not trigger review; a project must have sufficient federal involvement, such as federal funding, land, permitting, or approval. Where a privately developed data center depends primarily on state or local approvals—such as through MDEQ in xAI’s case—federal environmental review may never begin. Although it is difficult to establish the complete absence of federal involvement, records surrounding xAI’s Southaven development do not identify federal funding, federal land, or a project-specific federal approval that would clearly trigger NEPA review. Even when NEPA does apply, Robertson v. Methow Valley Citizens Council (1989) limits its force: agencies must take a “hard look” at environmental consequences, but NEPA does not require them to select the least harmful alternative. An agency may therefore acknowledge that a project worsens pollution or resource pressures and still approve it. Environmental review can make cumulative harm visible without making it legally decisive, leaving communities with a right to consideration, but not necessarily a right to prevent an additional burden.
Title VI provides a potential civil-rights check on environmental permitting, but its enforcement mechanisms are narrower than they first appear. Section 601 of Title VI of the Civil Rights Act, 42 U.S.C. § 2000d, prohibits discrimination based on race, color, or national origin in federally funded programs, and MDEQ’s receipt of EPA assistance brings its covered programs within Title VI; a state subsidy paid directly to a private company does not independently establish that federal-funding connection. Section 602 authorizes federal agencies to issue regulations prohibiting practices with discriminatory effects in federally funded programs, even in the absence of discriminatory intent, extending Title VI’s scope beyond intentional discrimination to encompass disparate impact at the regulatory level. However, in Alexander v. Sandoval (2001), the Supreme Court held that private plaintiffs cannot sue directly to enforce disparate-impact regulations under §602. A private §601 action requires intentional discrimination, meaning residents may demonstrate that permitting decisions disproportionately burden minority communities without being able to prove the discriminatory purpose necessary for a private claim. The gap between what Title VI makes legally relevant and what it makes privately actionable means that residents may demonstrate disproportionate burden without possessing an effective legal remedy—unless they can prove the kind of deliberate discriminatory intent that is rarely documented in modern permitting decisions.
The Clean Air Act’s citizen-suit provision gives affected communities a direct means of enforcing environmental law when public regulators do not. Section 304, 42 U.S.C. § 7604 expressly authorizes private plaintiffs to sue over specified violations after satisfying statutory prerequisites, while limiting suits when the government has already commenced and is diligently prosecuting an enforcement action. In the xAI litigation, the Department of Justice (DOJ) moved to dismiss the NAACP’s suit, citing Mississippi’s conclusion that the turbines did not require permits and the project’s asserted economic and national-security importance. Those policy concerns cannot alone defeat a cause of action Congress created; the DOJ must establish a recognized legal basis, such as lack of standing, defective notice, mootness, diligent government prosecution, or the absence of an underlying permit violation. Courts, rather than executive preference, must determine whether those requirements were met. That distinction matters for communities that rely on citizen suits when public enforcement is absent or politically constrained: preserving §304 protects their ability to enforce environmental law, even though it cannot create the broader cumulative-burden protections that statute presently lacks.
The limits of these federal statutes become clearer when compared with Monterey Park, California, where residents possessed authority over whether a data center could proceed or not. In June 2026, voters overwhelmingly approved Measure NDC, amending the city’s land-use law. The measure derives its legal force from Article XI, section 7 of the California Constitution, which permits cities to enact local regulations consistent with state law, and California Government Code §65800, which grants municipalities zoning authority. Although industry representatives warned that the measure would cost the city jobs, investment, and tax revenue—signaling that Monterey Park was “closed for business”—voters nevertheless concluded that those projected economic benefits did not outweigh the anticipated environmental and infrastructural costs. Unlike NEPA review or source-specific permitting, this authority changed the governing rule itself, preventing city officials from approving the prohibited use. Local zoning remains limited by possible state or federal preemption, the resources required for ballot campaigns, and the risk that development will shift to less politically powerful communities. Even so, Monterey Park shows what the federal framework often lacks: a mechanism through which cumulative community burdens can affect whether a project is approved at all, rather than merely how its individual impacts are reviewed or regulated.
The legal gaps point toward a three-part federal framework for AI infrastructure. Because NEPA under Robertson may disclose cumulative harm without requiring its prevention, federal review of functionally connected data-center infrastructure should incorporate cumulative-exposure analysis as a substantive factor in the permitting decision, rather than a disclosure requirement. Because Title VI under Sandoval does not generally provide a private right of action for disparate impact, the ability of affected residents to challenge permitting decisions that materially worsen environmental conditions in minority communities should not depend on proving deliberate discriminatory intent. And because the CAA regulates and enforces individual sources without determining when cumulative lawful pollution becomes unacceptable, agencies should possess authority to condition or deny permits when a project would materially worsen conditions in an already overburdened community. Under current law, a project may satisfy each individual legal requirement while no decision-maker is required to determine whether the combined burden imposed on the community should be permitted at all.
By Samantha Song
Section II: Ground Rules—Cost-Shifting Under the Federal Power Act
In Executive Order 14318, President Trump declared a “golden age” of American industry propelled by artificial intelligence (AI) data centers, which house various high-performance IT infrastructure used to train and run AI models, calling for “easing Federal regulatory burdens” to further accelerate construction. However, rapid construction of data centers across the country has simultaneously been met with a litany of legal challenges, many of which focus on the negative impacts data centers can have on surrounding residents. In particular, the extraordinarily high electricity demands associated with running a single data center create massive spikes in energy demand that strain existing energy supply chains. For example, xAI’s Colossus 2 data center carries an energy capacity of 1,563 megawatts, equaling the energy demand of over one million homes. Most of the increasing costs of sustaining data centers’ energy demand are shifted onto local communities, resulting in average increases of 267 percent in electricity prices in areas such as Virginia that have a high concentration of data centers. Currently, in line with EO 14318, no federal regulatory entity has adopted specific policies to address cost-shifting issues arising from data centers, despite its growing relevance to the future of U.S. energy markets. In the United States, the Federal Power Act (FPA) serves as the primary federal law governing interstate energy commerce, including electricity transmission and wholesale—sales between generators and utilities—energy rates. Part I of the FPA created the Federal Energy Regulatory Commission (FERC), which has jurisdiction to approve new interstate energy infrastructure and conduct independent investigations to ensure compliance with the FPA. The FERC and six Regional Transmission Organizations (RTO) under its authority regulate wholesale rates, which, through state-level utilities, influence the retail rates that consumers such as homes or businesses ultimately end up paying. As the United States’ primary electricity regulator, the FERC has already played a key role in accelerating the integration of large load customers, such as data centers, onto the grid. Utilizing its existing regulatory authority and obligations under Section 205 and 206 of the FPA, the FERC should establish clear market standards for wholesale electricity rates for preventing cost-shifting at the wholesale level, which would in turn protect retail consumers from bearing AI data centers’ large energy costs.
Recently, the FERC issued “show cause” orders to RTOs under Section 206 of the FPA to push them to “expedite the integration of large loads” onto the grid, aligning with EO 14318’s claim that rapid integration is key to maintaining a strong AI and innovation market for the U.S. While the FERC noted cost-shifting among potential concerns, they have maintained that the rise in energy demand and investment into new generation projects associated with data centers will be crucial in pushing the United States’ electric markets forward. To that end, the FERC has refrained from taking any targeted action to restrict or limit data center integration onto the grid, choosing to leave more authority to state-level utilities to account for regional and contextual differences. Similarly, heightened energy demand from data centers has also driven increased investment into renewable energy sources, potentially leading to a more sustainable grid in the long run. However, before those long-term benefits can be realized, cost-shifting remains a pressing and ongoing concern due to the disproportionate burden it places on consumers who have yet to experience substantial benefits from data center integration.
The FERC has yet to propose any additional rules or regulations specifically addressing the allocation of data center interconnection costs, prompting pushback and independent action from various state regulators. A statement submitted to the FERC by state ratepayer advocates from Maryland, Ohio, Delaware, and Illinois argued that the FERC’s latest orders inadequately address existing cost-shifting issues and proposed a rule that new customers should be required to internalize all of the costs of their connection. Independently, various state-level utilities have also begun to build similar rules into the legal framework. For example, in Pennsylvania, the Pennsylvania Public Utility Commission launched an investigation into PPL Electric Utilities Corporation over PPL inaccurately over-billing customers. The subsequently negotiated settlement (R-2025-3057164) aims to shield consumers from sudden rate increases by separating large-load customers such as data centers into a separate customer rate class, preventing them from sharing their own costs with other consumers. This settlement establishes rate class separation as a viable regulatory mechanism for managing cost-shifting issues that the FERC can similarly implement at the federal level. More broadly, multiple instances of action from state-level regulators demonstrate a clear need for targeted regulatory action to address cost-shifting issues, which the FERC has yet to address at the federal level.
Under the FPA, the FERC has a statutory obligation to maintain fair energy rates for all sales. Section 205(a) of the FPA mandates that rates and charges for all electricity sales under the FERC’s jurisdiction must be “just and reasonable” and prevents any public utility from maintaining any “unreasonable difference” based on different “localities or as between classes of service.” Cost-shifting, however, runs contrary to the FERC’s standards for reasonable rates, as it inequitably allocates higher costs to certain consumers. In the case of data centers, these higher costs also disproportionately affect nearby residents, creating an additional rate difference based on locality. Furthermore, the U.S. Seventh Circuit Court of Appeals found in Illinois Commerce Commission v. FERC (2014) that the FERC should allocate utility costs for prospective projects based on which consumers are likely to benefit from said projects, providing a clearer way for the FERC to determine unreasonable utility rates. The dispute in this case arose when PJM, the RTO for this region, proposed allocating the costs of new transmission lines and system upgrades equally across all of its utility members. However, because the new transmission lines were unlikely to benefit customers in the western part of the region, the Illinois Commerce Commission opposed the FERC’s proposed cost allocation. The Court held that utility projects should not shift “grossly disproportionate” shares of their costs onto other utilities that are unlikely to share their benefits, maintaining that everyday consumers should not have to pay for projects that do not benefit them. While EO 14318 emphasized the overall importance of data centers to U.S. manufacturing and technological power, a study from the University of Michigan found that data centers consistently left limited positive impacts on local communities, as most jobs created are term-limited and tax breaks primarily benefit corporate backers rather than local residents. These limited local benefits, combined with disproportionate energy cost increases for local residents, create the sort of imbalanced cost allocation the Seventh Circuit held the FERC is obligated to prevent. Under Section 205, the FERC is obligated to exercise its authority to prevent unfair electricity sale rates.
Furthermore, the FPA also obligates the FERC to play an active role in rectifying unreasonable sales. Under Section 206(a) of the FPA, if the FERC finds an existing rate to be discriminatory, preferential, or otherwise unreasonable, the FERC is required to determine and enforce a new and reasonable rate. One example from within the FERC’s existing regulatory framework for how it can approach the enforcement of reasonable rates is in its enforcement of grid reliability standards. Under FERC Order No. 2023, each RTO determines whether a new customer seeking to connect to the grid sufficiently meets the required reliability standards. If the new integration poses significant reliability risks, such as if it risks creating blackouts or grid instability, it is required to pay for the necessary transmission system upgrades, essentially requiring risky connections to internalize the costs of their connection. The FERC also has a legally enforceable penalty system for violators. In 2022, for example, the FERC brought a civil penalty of $4.4 million against PacifiCorp for failing to comply with its reliability standards. Just as the FERC requires risky grid connections to internalize upgrade costs under its reliability framework, it can use Section 206 to require large-load consumers like data centers to internalize the cost-shifting burdens they impose on surrounding ratepayers. In Fed. Energy Regulatory Comm'n v. Elec. Power Supply Ass'n (2016), the Supreme Court reinforced the FERC’s authority over rules and practices related to wholesale rates, particularly in the interest of maintaining “just and reasonable rates.” Though the FERC lacks direct jurisdiction over retail rates, which are what end consumers ultimately pay, the Supreme Court acknowledged that wholesale regulation from the FERC would unavoidably, albeit indirectly, impact retail rates. Though the FERC cannot directly regulate retail sales, it still has the authority to create standards and rules for wholesale rates that impact how customers internalize or pass on their costs. As a result, by drawing on both its existing legal authority and statutory obligation through Section 206, the FERC can effectively establish enforceable market standards to prevent cost-shifting onto consumers.
The FERC must take on a more direct role in creating stricter standards to ensure large-load customers such as data centers internalize their operating costs rather than shifting them onto surrounding communities. Under Section 205 of the FPA, the FERC is obligated to ensure just and reasonable rates for electricity sales, which have become increasingly threatened as the rapid integration of data centers onto the grid has spiked energy costs. The FERC’s consumer protection obligations under the FPA require stricter regulation of large-load integration, in contrast to EO 14318’s call for deregulation. Under the separation of powers, the FERC’s obligations under the FPA, a federal statute, would take precedence over an executive order. In cases where complying with EO 14318 conflicts with the FERC’s statutory obligations under the FPA, the FERC’s obligations under the FPA would prevail. As a result, the FERC is in a unique position to fulfill its statutory obligations under the FPA even where those obligations require a different approach than EO 14318 envisions for data center integration. In doing so, the FERC can provide a clear legal framework to address cost-shifting issues. A clear regulatory framework from the FERC requiring large-load customers to internalize their costs will provide a foundation for state-level utilities to build upon, ensuring that data centers do not compromise consumer protections while maintaining just and reasonable rates.
By Chloe Zhou
Section III: Losing Ground—Chemical Disposal Under the TSCA
The Toxic Substances Control Act (TSCA), signed into law in 1976 and substantially amended by the Frank R. Lautenberg Chemical Safety for the 21st Century Act in 2016, authorizes the U.S. Environmental Protection Agency (EPA) to maintain an inventory of commercial chemical substances, to require premanufacture notification before new chemicals enter commerce, and to restrict or ban any substance presenting an unreasonable risk of injury to health or the environment. For nearly five decades, the TSCA has governed how chemicals move through manufacturing, agriculture, and consumer products. More recently, however, a new category of industrial facility has emerged that concentrates regulated chemical substances at a scale few anticipated: data centers, the server farms that train and run artificial intelligence models, now consume fluorinated coolants, flame retardants, and battery electrolytes in quantities that rival traditional chemical-intensive industries. Although the TSCA does not mention data centers by name, the statute’s broad definition of chemical substances and its regulation of manufacturing, processing, and disposal bring data center operations within its existing statutory reach. While other regulatory frameworks address data center impacts through emissions permitting, environmental review, or energy allocation, the TSCA addresses a dimension those frameworks do not: the chemical substances that data centers use, store, and discard in the course of their operations.
The statutory text confirms that Congress wrote the TSCA to reach beyond the chemical landscape of 1976. Section 2(a) describes the statute’s scope as covering “chemical substances and mixtures which are constantly being developed and produced,” language that does not fix the TSCA to any particular era or industry but instead targets molecular identity regardless of industrial application. Similarly, the statute defines “chemical substance” as “any organic or inorganic substance of a particular molecular identity,” a definition broad enough to encompass substances that did not exist when the statute was implemented. Such breadth of language reflects congressional intent to create a regulatory framework capable of governing chemical risks as they emerge rather than only those known at the time of implementation.
Three categories of chemical substances used in data center operations satisfy this molecular-identity standard and trigger specific statutory obligations. First, data center liquid cooling systems use fluorinated gases and per- and polyfluoroalkyl substances (PFAS) as thermal management fluids. These are listed on the EPA’s TSCA inventory and are subject to ongoing EPA rulemaking efforts to restrict PFAS as a regulatory category, making the use of PFAS coolants in data centers directly relevant to one of the agency’s most active regulatory priorities. Second, backup power systems in data centers rely on lead-acid and lithium-ion compounds, both of which are TSCA-listed substances with established toxicity profiles. Third, server hardware contains flame retardants such as tetrabromobisphenol A (TBBPA) and chlorinated phosphate ester compounds, also listed on the TSCA inventory. Under Section 8(b), the EPA maintains an updated inventory of these chemical substances, and under Section 8(a), the EPA Administrator can require an entity that manufactures, processes, or distributes them to submit reports on their production volume, use, and exposure data. Because data centers use, process, and ultimately dispose of TSCA-listed substances, they are subject to these reporting obligations.
Data centers are not within the TSCA’s scope only because they use regulated chemicals; they also dispose of substances that the statute specifically governs, and disposal is where data centers pose the most direct environmental risk. During normal operation, chemical substances are largely contained within closed cooling loops and sealed battery systems. During disposal, however, those substances enter the environment through two different pathways.
The first pathway is the cooling tower blowdown. Data center cooling towers treat circulating water with borates, phosphates, and nitrites to prevent corrosion and bacterial growth. When these towers release concentrated water through blowdown, the substance-laden discharge enters municipal sewage systems or surface water. Among these substances, isopropylated phosphate is listed on the TSCA chemical substance inventory. While being listed on the inventory does not automatically impose specific restrictions, Section 6(a)(6) of the TSCA authorizes the EPA to prohibit or regulate “any manner or method of disposal” of any chemical substance that presents unreasonable risk. The EPA has not, to date, exercised this authority specifically for data center blowdown discharge, but the statutory mechanism exists, and the listed substances are present. The TSCA does not directly regulate discharge into waterways, an authority that belongs to the Clean Water Act, but it regulates the substances themselves before and during disposal, creating a complementary regulatory hook that the Clean Water Act alone does not provide.
A second disposal pathway arises when servers reach the end of their operational life. Server hardware containing TBBPA and chlorinated phosphate ester compounds is discarded, and these substances can leach into soil and groundwater through landfill disposal or be released into the atmosphere through improper incineration. Section 3 of the TSCA covers substances whose “manufacture, processing, distribution in commerce, use, or disposal may present an unreasonable risk of injury to health or the environment,” and end-of-life server disposal falls squarely within this language. In both disposal pathways, Section 6 gives the EPA authority to restrict disposal methods that present unreasonable risk, making data center disposal practices subject to potential EPA regulation under the TSCA even where the EPA has not yet exercised that authority.
The TSCA was designed to regulate chemical substances without impeding technological progress. Section 2(b)(3) explicitly provides that the statute’s regulatory policy should not “impede unduly or create unnecessary economic barriers to technological innovation.” Applying the TSCA to data centers does not conflict with this principle. Regulating how data centers handle and dispose of PFAS coolants, flame retardants, and cooling tower chemicals does not prevent data centers from being built or artificial intelligence models from being trained. It ensures that the chemical substances involved in those operations are handled safely, the same way the TSCA ensures chemical safety in manufacturing, agriculture, and every other industry that processes regulated substances.
The gap, however, is not in the TSCA’s statutory reach but in its enforcement. Data center operations involve the use and disposal of TSCA-listed chemical substances, bringing those operations within the statute’s regulatory scope. To date, however, the EPA has not publicly interpreted the TSCA as applying to data center operations, initiated enforcement actions against data center operators for chemical substance violations, or incorporated data centers into its ongoing PFAS rulemaking despite the industry’s growing use of fluorinated coolants. This enforcement gap exists because the agency has not yet applied the TSCA and its subsequent authority to a rapidly emerging industrial category. In 2016, the Lautenberg Act amendments strengthened the EPA’s authority by requiring risk evaluations for existing chemicals and imposing deadlines for EPA actions, which could be directed toward data center chemical use if the agency chose to prioritize it. As data centers continue to expand and their chemical consumption grows, the question is not whether the TSCA covers these operations—the statutory text confirms that it does—but whether the EPA will exercise the authority Congress has already granted.
By Jennie Ren
Edited by Jemma Granite and Sylvia Martínez-López
This piece was reviewed and finalized by Qizhen (Kiara) Ba.
The views in these articles are those of the individual authors and not of the Columbia Undergraduate Law Review.