U.S. Senator Bernie Sanders of Vermont, in announcing his support for a congressional bill stipulating a data center moratorium, named generative artificial intelligence (AI) “the most consequential technology” of all of humanity.Footnote 1 The lack of federal constraints on data center expansion in the United States has allowed the technology industry to invest significantly in the infrastructure necessary to power generative AI models, with nearly 3000 new data centers under construction or planned across the country.Footnote 2 The companies championing this growth are colloquially referred to as “hyperscalers,” or platform operators that are driving the development of large language models and scaling cloud computing services globally.Footnote 3
We find it important to bring into conversation the use of generative AI—its purpose and organizing logics—with the infrastructure it necessitates. AI technologies have automated and replicated measures that murder, criminalize, surveil, and replace Black, Brown, Indigenous, and poor people around the world.Footnote 4 Hyperscale data centers and their power facilities mark a new chapter in the history of extractive industries that enact deadly conditions in communities across the United States.
Ruha Benjamin’s call for a “race-conscious orientation” is critical to understanding these technological assemblages and to wayfinding toward a human(e) future.Footnote 5 Hyperscale data centers in the United States are reproducing conditions that exacerbate risk and burden for Black, Brown, Indigenous, and poor communities, leading to premature death.Footnote 6 This reproduction of harm follows the logic of racial capitalism—which underlines how capitalism necessitates racial hierarchies of difference to reproduce extraction. Through this analysis, infrastructure is both a tool for reproducing and a product of social control.Footnote 7 Extraction does not only occur at the site of the data center; it is an assemblage of the “specific processes, logics, ideologies, and relations that recast lands, labors, ecosystems, and bodies” as expendable for wealth accumulation.Footnote 8 In the current data center landscape, imperatives that center capital and render marginalized bodies disposable are actively shaping political and economic life in the United States. We have seen this already in the disproportionate burdens faced by marginalized communities that host hyperscale data centers, including unaffordable utility bills and the defunding of public education and other critical services that support social reproduction in favor of AI innovation. We need to develop liberatory data futures, moving at what co-author Taeya Boi-Doku names a “more human pace.”
1. Compounding burdens on poor, communities of color
The unequal apportionment of livable environments—clean air and affordable, non-lethal electricity sources—is central to the functioning of extractive assemblages. This burden is most notable with the development of Colossus, a pair of hyperscale data center facilities in the neighboring cities of Memphis, Tennessee, and Southaven, Mississippi, and the associated cluster of natural gas-powered turbines used to power the data center cluster. Colossus is owned by xAI, a hyperscale AI company developing the large language model Grok, championed by Elon Musk. The construction of Colossus is a new page in the history of extractive infrastructure in the area. Boxtown, a predominantly Black neighborhood in Memphis, is the second worst in the nation for asthma-related emergency doctor visits, and the seventh in asthma-related deaths.Footnote 9 This abnormal rate of illness has been linked to the overwhelming presence of oil refineries and power plants proximate to the community. Since natural gas turbines were brought online for Colossus, residents have reported worsening asthma attacks, lingering smells of noxious gas, and an inability to breathe.Footnote 10
xAI is being sued by Earthjustice, the Southern Environmental Law Center, the National Association for the Advancement of Colored People (NAACP), and the Mississippi State Conference of the NAACP, over its illegal operation of 27 gas turbines that have the potential to emit more than 1700 tons of nitrogen oxide, 180 tons of particulate matter, 500 tons of carbon monoxide, and 19 tons of formaldehyde per year, chemicals known to cause cardiovascular disease and cancer at such exposure levels.Footnote 11 Since the original writing of this article, the Department of Justice has filed to dismiss this lawsuit, citing that national security concerns, of which include AI development for defense, should take precedent over environmental regulation.Footnote 12
The race to scale large language models incentivizes AI developers to site their development in localities where there will be a path of least-resistance politically and economically, exacerbating existing vulnerabilities. The current pace of hyperscale data center development secures a future of continued reliance on natural gas and fossil-fuel–powered electricity, which has been found repeatedly to cause deadly conditions in marginalized communities.Footnote 13 Low-income, rural, communities of color have long been deliberate targets of the fossil fuel industry, as environmental justice literature amply demonstrates—and data center developers are showing early indications of repeating this pattern.Footnote 14
In addition to grave health impacts, energy affordability is a key concern that host communities are already experiencing. In the largest cluster of hyperscale data centers in the world, Loudoun County, Virginia, an independent study by the state government found that due to electricity demand driven by new data centers “a typical residential customer could experience [electricity] generation- and transmission-related costs increasing by an estimated $14 to $37 monthly in constant or real dollars by 2040, independent of inflation.”Footnote 15 Energy burden is considered significant when 6 percent or more of a household’s income is directed toward energy costs for subsistence services, such as utilities.Footnote 16 This burden has been found to be distinctly racialized, even when controlling for income. African American majority census tracts tend toward higher rates of energy burden, due to histories and contemporary experiences of redlining, racist financial exclusion, and gaps in federal assistance available to renters.Footnote 17 This means that the increasing strain from data centers on the grid is forcing many households to choose between foregoing electricity, medication, or childcare services.Footnote 18
2. State deregulation, or privatization of public services and subsidization of private needs
The state’s active and explicit role in permitting data center development generates conditions of lethality for marginalized communities. The second Trump Administration has purposefully hollowed out its own capacity to regulate, refused to enforce health and environmental statutes, and defunded bureaucratic expertise that could have otherwise stalled or slowed development.Footnote 19
The thousands of concurrent data center developments across the nation would not be possible at this speed and scale were it not for the federal government’s deliberate support for streamlining permitting processes and refusal to enforce air pollution standards set by the Clean Air Act.Footnote 20 The Colossus data centers use a natural gas–powered shadow grid that is operating without Clean Air Act permits.Footnote 21 Despite independent regulators at the Environmental Protection Agency (EPA) finding the actions by xAI to be illegal and in breach of the Clean Air Act, public health officials in Memphis and lead EPA administrator Lee Zeldin are circumventing environmental regulations to ensure that the xAI project continues.Footnote 22 The lack of government intervention at these toxic energy sites is a clear sign of the re-rationalization that creates disposable life and abandons community interests in favor of technocratic profit interests.
No state government or public utility commissions, at the time of writing, has legally responsibilized data center operators for the load requirements and grid infrastructure updates that their energy demand necessitates to date.Footnote 23 Instead, utilities have spread the capital costs of new private power infrastructure projects onto ratepayers who will not reap the benefits of the grid’s increased energy capacity.Footnote 24 Here, utilities are drawing upon ratepayer funds to subsidize the real cost of data centers’ electricity demand.Footnote 25 Compounded with the patchwork of differentially extended electrical networks to communities of color, this could lead to extreme variations in the cost of basic utilities in the United States.Footnote 26
Public opposition to this system is strong, and data center operators are already building out off-grid energy plants known as “shadow grids” to avoid potential future regulation that may require them to bear the true cost of their power. Data center projects can be delayed for years due to the grid connection, due to lengthy permitting processes and grid connection wait times.Footnote 27 Thus, these behind-the-meter energy plants serve a dual purpose—generating electricity using natural gas systems that do not require grid interconnection.Footnote 28 This approach maximizes both profit and speed, despite natural gas processes that are wildly inefficient, costly, and deadly in the short-term, which include strapping mobile gas generators to semi-trucks, originally designed for aircraft or cruise ships.Footnote 29
The question of water use also reveals the aligned interests of data center operators, utilities, and the state. While estimates of data centers’ gross water consumption vary due to the lack of government-mandated corporate transparency, Han et al. found that US data centers could collectively require “697–1,452 million gallons per day of new water capacity through 2030, comparable to New York City’s daily average” if 2024 water-use intensity persists.Footnote 30 For water-stressed areas, such as the Southwestern United States, data center water usage has likely exacerbated drought conditions. Just two years after multiple data center campuses came to the region, Arizona experienced one of the worst droughts on record.Footnote 31 As Garcia notes, “Farmers were forced to fallow fields and some communities outside of Phoenix even went without tap water due to the Colorado River dwindling.”Footnote 32 Datacenter operators, when prompted to report water usage rates after an inquiry to the city was issued, redacted exact figures because the data were considered “proprietary.”Footnote 33 A similar case occurred in The Dalles, Oregon, where a request for transparency from the local data center on its water usage, led to a hyper scale company suing to prevent their water records from becoming public.Footnote 34 Given the legacy of environmental racism in Flint and the settler colonial enclosure of water that inhibits Tribal sovereignty, how can communities even attempt to adapt to deadly conditions when data center operators, in alliance with the bureaucratic contours of the state, refuse to disclose their water usage?Footnote 35
Despite the ample documentation of data centers reproducing environmental violence, deadly conditions, and exacerbating rate costs, they are still desired by state and local officials. Thirty-eight states and a range of municipalities across the United States offer tax abatements and subsidies to data center operators.Footnote 36 Narratives promising new jobs and economic revitalization are circulated to gain community consent for development, aligning with past fossil fuel campaigns that exploited precarity to garner local support for detrimental infrastructure.Footnote 37
Such promises have gone un- and under-realized, with data centers creating jobs during peak construction and minimal technical and security staffing throughout the data center’s operational lifetime.Footnote 38 Texas will reportedly lose $3.2 billion in sales tax revenue over the next two years due to its current exemption program for data center operators.Footnote 39 When the state is willing to lose valuable dollars, what programs bear the cost of those lost funds? In Michigan, this loss of significant revenue was felt in the public education budget, teachers’ salaries, and facility maintenance due to the Switch data center’s “exemptions from property taxes that funded school districts.”Footnote 40 Instead of receiving investments from the data center industry to support local school districts, “Michigan lawmakers initially considered legislation that would have required school districts to reimburse the company for taxes already paid, further straining school finances.”Footnote 41 Is this what it means for a data center to act as a good neighbor?
3. A more human pace
Silicon Valley’s motto is to “move fast and break things.” Generative AI, and the hyperscale data centers they require, are doing just that.
To move at a human(e) pace is to heed the call of Sylvia Wynter and Katherine McKittrick, to move beyond the colonial and racial orderings of knowledge and logic that have allowed this technology to proliferate, and to re-understand what our purpose as human beings is in relation to each other and the world.Footnote 42 This requires a move away from the enclosure of public information, human creativity and ingenuity, and the further accumulation of wealth through the dispossession of people from their right to a livable, desirable life.
We acknowledge the important work of frontline communities that are attempting to negotiate community benefits agreements and ordinances as a form of resistance. Communities have been successful in negotiating funds for public education, home weatherization, and digital literacy.Footnote 43 Where negotiation is not desired by the community, “coordinated local opposition, moratoria, and litigation” have seen upward of $156 billion dollars wasted in blocked or delayed projects.Footnote 44 Both the potential for negotiation and the persistent loss of upfront investments strengthen the value of collective organizing through projects like the NAACP’s Frontline Framework Community Data Center Guiding Principles, which provide scaffolding for communities to advocate for data centers that can contribute to community benefits and environmental justice.Footnote 45
Data center expansion is deeply emergent, and we call on research to support the language, study, and multi-year inquiries needed to support the widening of the political imagination toward liberatory data futures. Given the assemblage in which data centers currently operate and the nearsighted strategy of the fossil fuel industry that they seek to reproduce, there is no scenario in which AI data centers can contribute to non-extractive futures. Sneha Krishnan reminds us that we can build solidarity through “being the killjoy, the complainer, the one who persistently says no and stalls the work” of AI infrastructure and its interlocutors.Footnote 46 Here, Sanders’s moratorium becomes the imperative for all who long for breathable, affordable, and dignified futures. Delays, setbacks, stalemates, and limited opportunity for siting are pathways to unlocking the potential of data centers as human(e) assemblages.
Author contribution
Conceptualization: A.D., S.L., T.B.; Data curation: S.L.; Formal analysis: A.D., S.L., T.B.; Investigation: A.D., S.L., T.B.; Methodology: A.D., S.L., T.B.; Project administration: A.D., S.L., T.B.; Resources: A.D.; Validation: A.D., S.L., T.B.; Writing - original draft: A.D., S.L., T.B.; Writing - review & editing: A.D., S.L., T.B.
Conflict of interests
The authors are colleagues at an environmental justice nonprofit organization, which conducts strategic grantmaking initiatives that support workforce development, renewable energy deployment, and community organizing. This organization did not finance or contribute to this work.