The Case for Heightened Scrutiny: AI Development, Pipelines, and Eminent Domain

Michael Hren

Edited by Keerthi Chalamalasetty, Eshal Charolia, Mac Kang, and Sahith Mocharla

Ever since OpenAI released its first LLM in 2022, the world has been swept up in an AI frenzy. In six years, the New York Stock Exchange’s market capitalization doubled from 22 to 44 trillion, largely driven by the AI boom [1]. Yet, AI-fueled economic growth depends on the construction of data centers across the United States and around the world; however, this growth comes at the expense of local environmental costs. A report by the Brookings Institute shows that a single large-scale artificial intelligence (AI) data center can require millions of gallons of water daily for cooling and operations, placing severe strain on regional water supplies that may already be depleted by climatic factors [2]. In Quincy, Washington, a small town along the Columbia River, the local public utility district seized 84 properties to build electrical infrastructure for new AI data centers, ultimately doubling local residents’ electricity costs [3]. 

Quincy’s public utility district derives the power to do so from the principle of eminent domain. In Kohl v. United States, the Supreme Court affirmed the federal government’s power of eminent domain to seize private property for public use [4]. Specifically, Kohl ruled that the U.S. government had the right to seize private land near Cincinnati to build a post office [5]. As the post office’s public function depended on its proximity to the city, it had to be there for the public good. However, this definition has expanded to allow private enterprise to seize property, provided that public interest and just compensation are provided. In Quincy, this leads to the public utility commission seizing land for private companies’ projects. Unlike traditional infrastructure, such as electric grids, highways, or government buildings, which must be located in a particular area to fulfill their function, AI data centers, for the most part, are far more flexible in location. Although residents and public utility departments recognize that data centers could theoretically contribute to the public good, their construction has immediately led to water loss, strain on electrical grids, and higher electricity costs for consumers. Fundamentally, AI infrastructure differs from the types of infrastructure for which eminent domain was designed. Thus, the current eminent domain framework is ill-equipped to handle the burgeoning data center industry. Clearly, data centers must meet a different and higher burden of justification. 

Traditional eminent domain doctrine developed around projects that required specific locations for the public good, giving the government broad leeway to seize land. This principle was reaffirmed in Berman v. Parker, where the Court ruled eminent domain can be used to achieve a public purpose, such as the beautification of a slum [6]. In 1945, Congress passed the District of Columbia Redevelopment Act, which granted the power of eminent domain to identify and redevelop neglected areas of (southwestern) Washington, D.C. [7]. Critically, Berman expanded the definition of eminent domain from a specific public use to a broader, congressionally defined public purpose. Moreover, Justice William Douglas writes for the majority that, “subject to specific constitutional limitations, when the legislature has spoken, the public interest has been declared in terms well-nigh conclusive [8].” In Quincy’s case, Congress has already framed AI as a critical public resource to be harnessed, thereby making takings justifiable under existing legislation and Constitutional precedent [9]. 

Although Berman expanded the doctrine to encompass a public good and empowered Congress to define what constituted such a good, the DC redevelopment plan was inherently location-specific; building elsewhere would not satisfy Congress's aim of redeveloping Southwest DC. Overall, both Kohl and Berman shared the assumption that the project’s location was non-negotiable. Accordingly, eminent domain doctrine in the first half of the 20th century rested on geographic proximity and the measure of necessity for a property, as key factors in determining whether a property meets the threshold for eminent domain. That assumption, though intuitive, would not survive the next half-century.

In 2005, the Supreme Court, in Kelo v. New London, broadened the definition of public use to encompass economic development projects carried out by private entities. [10]. The city of New London, Connecticut, used its power of eminent domain to seize private property to sell to developers as part of a broader economic revitalization plan. [11]. Furthermore, during the case, New London did not provide any evidence that the private contracts served a public purpose; instead, it cited the project’s development plan. While the plaintiffs argued that contracting with private developers did not constitute public use, the Court ruled in favor of New London, concluding that the intent to create jobs and economic development constituted a sufficient public purpose. In doing so, the Court affirmed that theoretical economic growth can constitute a public good even if the immediate beneficiary is not society. In Quincy, this led to land seizures that disproportionately helped private companies, while disadvantaging the public.

Private companies, though, do not inherently possess eminent domain power; states must explicitly delegate it to public utility companies or enact legislation. The precedent established in Kelo v. New London permits the government to justify takings for AI data centers on the grounds of economic development, even when the primary economic beneficiaries are the AI company and its investors, as opposed to general or local society. Under Kelo alone, that is sufficient. Yet, although Kelo is codified at the federal level, it has not preempted state law, allowing states to set higher burdens. Permitted state-level resistance to Kelo demonstrates that the Supreme Court and state legislatures recognized that Kelo’s logic was overly permissive. In 2005, Texas, for example, passed SB7 to prohibit economic development or tax revenue from being constituted as public use under the takings clause [12]. In 2006, Florida passed HB 1567 (2006), which mandates a ten-year waiting period for eminent domain transfer to a private entity [13]. Overall, many states have imposed a higher burden on private companies when exercising eminent domain, leaving only slight exceptions for private use under the takings clause of the Fifth Amendment. For the town of Quincy, state pre-emption over Kelo provides a statutory precedent for establishing higher barriers for AI takings.

The precedent for eminent domain was established for infrastructure that needed to be constructed in a specific geographic location: highways, housing projects, and government buildings. By contrast, AI data centers can be built anywhere open land is available, provided they have sufficient connectivity and power [14]. A limited exception exists in the application of edge computing, which requires close geographic proximity to population centers, but does not represent the vast majority of data center construction [15]. This reasoning is supported by the current administration’s own proposals, which suggest placing AI data centers on Superfund sites near every major American population center [16] [17]. 

Such proposals implicitly acknowledge the geographic flexibility of AI data centers, raising an important question: Does infrastructure that is not geographically constrained satisfy existing justification for eminent domain, particularly when its placement imposes localized environmental and economic harms? 

Quincy, Washington, illustrates this tension. Private developers and utility companies effectively chose where to build data centers and related infrastructure. But the location-induced consequences, on the other hand, including doubling electricity costs and diminished water quality, are sociophysical consequences that local residents must bear [18]. AI data centers can be located on degraded or remote land, such as Superfund or brownfield sites [19]. While companies need to clear various legal hurdles under the Comprehensive Environmental Response, Compensation, and Liability Act (CERCLA) and Environmental Protection Agency (EPA) guidelines, the administration’s proposal acknowledges AI’s flexibility in location [20]. Thus, the proximity of current AI data centers to water-scarce areas seems more like an argument for convenience or efficiency than a public-use justification required to exercise eminent domain.

Although AI projects benefit from Kelo’s public use standard, this has enabled takings that have led to obvious/clear, and significant negative repercussions for local communities. However, prior disputes over energy and water use have established precedent supporting stronger protections for projects that threaten public water and energy resources. In PUD No. 1 of Jefferson Cty. v. Washington Dept. of Ecology, the Supreme Court affirmed Washington’s authority under the Clean Water Act to require minimum stream flows as a condition for building a hydroelectric dam [21]. Even though the Clean Water Act only permitted protection in areas of water quality, such as pollution levels, the Supreme Court established stronger protections against infrastructure projects because they threatened water quantity [22]. The Court reasoned that because water quantity is a part of water quality, the state could condition a permit to maintain a minimum flow [23]. Two decades later, in State v. Cooper (2015), the New Hampshire Supreme Court awarded the state damages against Exxon for groundwater contamination [24]. Although New Hampshire was not directly affected, the Court reasoned that the public trust in water resources implicated them [25]. While neither case arises in the eminent domain context, both reflect a willingness to impose heightened regulatory burdens when projects threaten shared public resources. There is no principled reason why this logic should not extend to eminent domain, where the public use justification itself depends on net public benefit. While this ruling would come too late for Quincy, which has already experienced significant development, this higher standard could prevent other towns from becoming resource-strained like Quincy.

Given their relative flexibility in location and substantial impact on water resources, AI data centers should be required to meet a higher burden of justification before invoking eminent domain. The constitutional foundation of eminent domain rests on serving the public good [26]. Therefore, when a project places a grave strain on public resources and impacts public health, it should not qualify as a public good or, at the very least, should meet a greater threshold to qualify as a public good than normal. 

Existing regulatory frameworks illustrate how heightened burdens are already imposed across comparable contexts. For instance, interstate natural gas pipelines seeking to exercise eminent domain must obtain a certificate of public convenience and necessity from the Federal Energy Regulatory Commission (FERC) under the Natural Gas Act [27]. To obtain this, developers must go beyond basic eligibility requirements to demonstrate that the project serves public convenience and necessity [28]. Similarly, large mining operations that discharge waste into US water must obtain permits under Section 404 of the Clean Water Act [29]. To obtain this, projects must undergo strict public review and meet mitigation requirements to address potential environmental impacts [30]. In Coeur Alaska v. Southeast Alaska Conservation Council (2009), Coeur Alaska Inc. wanted to reopen a gold mine in Alaska and obtained a section 404 permit from the Army Corps of Engineers [31]. However, the Southeast Alaska Conservation Council argued that Coeur Alaska should instead be required to obtain the EPA’s stricter Section 402 permit, as the mine not only took away common pool resources, but regulators recognized the project’s potential impact on public goods through discharge and construction [32]. While the Supreme Court ultimately sided with Coeur Alaska, this case demonstrates the ongoing debate over imposing higher burdens on projects that threaten public environmental goods.

Together, these frameworks show that private developers must demonstrate public necessity before exercising eminent domain. However, unlike pipeline or mining developers, there is no suitable equivalent barrier before the government condemns land for private AI infrastructure, despite similar circumstances and impacts.

From Kohl to Kelo, the Court has invoked geographic necessity to justify deference. Data centers, from the government's own publications, do not meet this assumption while imposing significant resource burdens on water and energy systems, concerns that Congress has already recognized as warranting higher scrutiny in other contexts. Without a higher burden for establishing eminent domain, the local human costs of AI-related projects will far outweigh the public good. Both the courts and Congress have acknowledged the potential dangers of AI to the population and have instituted higher burdens for similar projects that threaten public resources. Therefore, legal reform in this instance is both plausible and consistent with existing doctrine and necessary. Congress, for example, could institute a FERC-style certification schema before a data center can exercise eminent domain in water-sensitive areas. As AI infrastructure expands, ensuring that eminent domain serves public benefit rather than convenience or efficiency will determine whether it remains faithful to its constitutional foundation or becomes a tool for private enterprise.

---

[1] World Fed’n of Exchanges, Market Statistics, WFE FOCUS (Jan. 2020), https://focus.world-exchanges.org/issue/january-2020/market-statistics.

[2] Why Water Is Important for Data Centers | The TechTank Podcast, Brookings, https://www.brookings.edu/articles/why-water-is-important-for-data-centers-the-techtank-podcast/ (last visited Apr. 1, 2026).[3] The Impacts of AI Data Centers in Rural Communities: Hearing Before the Subcomm. on Rural Development, Energy, and Supply Chains of the H. Comm. on Small Business, 119th Cong. (2026) (testimony of Camilla Taylor, American Sustainable Business Network) .[4] Kohl v. United States, 91 U.S. 367 (1875)[5] See [4][6] Berman v. Parker, Oyez, https://www.oyez.org/cases/1940-1955/348us26 (last visited Apr 1, 2026).

[7] See [6]

[8] See [6][9] H.R.4346 - 117th Congress (2021-2022): CHIPS and Science Act. (2022, August 9). https://www.congress.gov/bill/117th-congress/house-bill/4346[10] Kelo v. New London, Oyez, https://www.oyez.org/cases/2004/04-108 (last visited Apr 1, 2026). [11] See [10][12] House Research Org., Bill Analysis, Tex. H.B. 16, 79th Leg., 2d C.S. (2005).[13] Florida House of Representatives - Documents, https://www.flhouse.gov/Sections/Documents/loaddoc.aspx?chapter=0073&section=013&DocumentType=StatRev&PublicationType=S&BillId=33829 (last visited Apr. 1, 2026). [14] Konstantin Pilz & Lennart Heim, Compute at Scale: A Broad Investigation into the Data Center Industry (Nov. 22, 2023), http://arxiv.org/abs/2311.02651. [15] Whitepapers & Reports – LF EDGE: Building an Open Source Framework for the Edge., https://lfedge.org/resources/whitepapers-reports/ (last visited Apr. 1, 2026). [16] Accelerating Federal Permitting of Data Center Infrastructure, The White House (July 23, 2025), https://www.whitehouse.gov/presidential-actions/2025/07/accelerating-federal-permitting-of-data-center-infrastructure/. [17] How Close Are You to a Superfund Site?, http://www.nationalgeographic.com/superfund (last visited Apr. 1, 2026). [18] See [3][19] See [16][20] U.S. Envtl. Prot. Agency, Guidance on the Redevelopment of Superfund and Brownfield Sites as AI Data Centers, EPA-540-S-26-001 (Jan. 2026), https://www.epa.gov/superfund/guidance-redevelopment-superfund-and-brownfield-sites-ai-data-centers.

[21] PUD No. 1 of Jefferson Cty. v. Washington Dept. of Ecology, 511 U.S. 700 (1994) 

[22] See [21][23] See [21]

[24] State v. Cooper, 168 N.H. 161 (2015). [25] See [23][26] https://scholarship.law.cornell.edu/cgi/viewcontent.cgi?article=3043&context=clr (last visited Apr. 1, 2026). [27] Interstate Natural Gas Pipeline Siting: FERC Policy and Issues for Congress (2026), https://www.congress.gov/crs-product/R45239. [28] see [27][29] OW US EPA, Permit Program under CWA Section 404, (Mar. 17, 2015), https://www.epa.gov/cwa-404/permit-program-under-cwa-section-404. 

[30] OW US EPA, Overview of Clean Water Act Section 404, (Mar. 3, 2015), https://www.epa.gov/cwa-404/overview-clean-water-act-section-404. [31] Coeur Alaska, Inc. v. Southeast Alaska Conservation Council, 557 U.S. 261 (2009)[32] See [31]