Study finds racial disparities in US data center locations | Magica
Study finds data centers more likely in Black, Hispanic and Asian neighborhoods
Editorial Team
••📖7 min read
A national study finds higher odds of hosting data centers in majority-Black, Hispanic and Asian neighborhoods. Facility types, zoning rules and unequal access to public hearings complicate who bears the costs and receives the benefits.
A nationwide study finds majority-Black, Hispanic and Asian neighborhoods are more likely to host data centers than majority-white neighborhoods.
Researchers found racial disparities persisted after accounting for economic and infrastructure factors.
Location, access to public hearings and the distribution of potential economic benefits emerge as distinct questions.
A nationwide study finds that America's data centers are disproportionately located in communities of color, while the large technology-company campuses associated with greater potential local economic benefits are more likely to be in majority-white neighborhoods. The findings put facility location, potential returns and residents’ ability to negotiate before approval at the center of the infrastructure boom.
Odds of hosting a data center relative to majority-white neighborhoods in the same state. These ratios do not measure absolute probability or pollution exposure. Source: 404 Media.
A national pattern beyond individual projects
Researchers at George Washington University Law School and American University examined roughly 81,000 US census tracts and the locations of 5,291 data centers, according to the account of their analysis. These facilities house computing and networking infrastructure. The study addresses data centers broadly; its facility count should not be read as a tally of installations devoted exclusively to AI. Census tracts provide its neighborhood comparison.
The reported comparisons are odds ratios between neighborhoods in the same state:
Neighborhood majority
Odds of hosting a data center relative to majority-white neighborhoods in the same state
Epic says it is using AI to rewrite parts of Unreal Engine and plans to let developers connect their preferred models to UE6. The strategy combines claimed maintenance gains with a broader overhaul of programming, distribution and portable game assets, while rival Unity already offers outside-model connections.
Editorial Team
Odds compare the chance that a neighborhood hosts a facility with the chance that it does not. These ratios are not multipliers of absolute probability, and they do not measure an individual resident’s exposure to pollution.
A separate descriptive comparison in coverage of the research puts the average share of residents who are people of color at 44% in data center neighborhoods, compared with 38% in neighborhoods without them—a six-percentage-point difference. Those averages describe neighborhood composition, rather than the share of all Americans living near facilities.
The researchers reported that disparities persisted after accounting for population, household income, education, land costs, proximity to major transmission lines and differences between states. That matters because data centers cluster around metropolitan areas, which are often more racially diverse. But statistical adjustment does not establish the motive behind a particular siting decision or measure the harm caused by each facility.
The study, “Redlining in the Age of AI: The Relationship between Data Center Locations and Race in the US,” was co-authored by Samantha Bradshaw, executive director of American University's Center for Security, Innovation, and New Technology, whose earlier research focused on disinformation and democracy.
Co-author Dhanaraj Thakur is director of the Fair Technology Initiative at George Washington University Law School. He previously researched election disinformation as research director at the Center for Democracy & Technology. A 2021 report he co-authored examined that problem in the United States, Brazil and France.
The authors said reporting on data centers polluting Black and Brown communities in Memphis prompted them to test whether the pattern extended nationwide.
Bradshaw also pointed to Silicon Valley and Northern Virginia's Fairfax and Loudoun counties as places where concentrations of Asian Americans overlap with data center development. The national disparity does not mean every community has the same development history, facility type or exposure.
Existing zoning can limit public hearings
The historical argument is about how earlier decisions shape today's available sites. Redlining denied access to credit based on where people lived, particularly in Black neighborhoods. A Federal Reserve history explains that the federal government helped institutionalize the practice; racially motivated redlining was outlawed by the Fair Housing Act in 1968.
The researchers connect that legacy of segregation to the placement of industrial land and power infrastructure. Transmission lines make sites attractive to electricity-intensive facilities; the authors argue that their historical placement helps explain today’s racial disparities. This historical account does not independently establish the cause of any particular siting decision.
Thakur said he doubted developers were intentionally seeking communities of color. His explanation centers on race being embedded in other siting factors, rather than proof of deliberate targeting by each developer.
As reporting on the approval process explains, a project requiring rezoning generally faces public notice, a hearing and a vote. A project on land already zoned for industrial use can instead proceed through administrative review without those opportunities for intervention.
Separate tracking by Zoned Out, a tool created by Harvard researchers, illustrates the stakes. In the projects it tracked, data centers were blocked 66% of the time when communities received a hearing; without one, 80% were approved. Communities receiving hearings averaged nearly 80% white, while those without hearings were roughly 60% people of color, according to the same reporting. The cited account does not give the project count or observation period behind those percentages. They describe the tracked projects, rather than a demonstrated national probability of approval. They also do not prove that a hearing itself caused rejection: the projects and communities entering each approval route may differ.
The same reporting describes a developer withdrawing a proposal after opposition in a wealthy, 65% white North Carolina community, then moving it weeks later to another county. The closest neighborhood at the new location was predominantly Black and had a median income one-third as high; county commissioners approved the project 6–1. That sequence illustrates how development can move after opposition, without establishing a universal developer strategy.
In Augusta, Georgia, resident Zakiya Mabery learned about the $2 billion Project Eisenhower after officials had approved it. The 200-acre project was first authorized in 2022 without a public hearing and began construction in 2026. The closest community is 67% Black. A subsequent temporary moratorium on data center development did not cover that project.
The pressure extends beyond local zoning. The same reporting describes the Trump administration’s push to speed construction by changing federal environmental review rules and limiting opportunities for public comment on data center projects. Federal review and local land-use approval are distinct decisions, but both determine when residents can intervene.
The economic promise depends on the facility
The study distinguishes large campuses, often called hyperscale facilities, from smaller colocation facilities that can serve as both network hubs and computing sites. Counting every location as equivalent obscures what a host community might receive.
That distinction complicates the industry argument that hosting infrastructure delivers jobs and tax receipts. According to the reported findings, campuses operated by technology companies such as Meta, Google and Microsoft are more likely to generate property-tax revenue and implement community benefit agreements. Those campuses are also more likely to be in majority-white neighborhoods.
Thakur cautioned that industry promises of jobs and tax receipts are themselves contested; even accepting those promises, the research suggests communities of color have less access to the potential rewards. The reported siting pattern does not establish how much money ultimately reaches households living nearest a facility.
The geography also needs care. The separate reporting notes that the two closest towns to Meta’s northern Louisiana project are more than 80% Black. A national tendency about host neighborhoods does not describe every nearby town, and neither tells readers who gets hired.
Bradshaw raised a further complication: revenue does not always stay in the host neighborhood or county and can be distributed at the state level. A facility's location and its beneficiaries therefore need to be examined separately. Noise, water demand and concerns about utility bills are local issues even when the economic gains flow more widely.
Political visibility is uneven, too. According to the account of the researchers’ findings, communities with media-recorded opposition averaged about 79% white, compared with 53% in communities without recorded opposition. That measures documented resistance, rather than whether other residents consented. The authors interpret the gap as reflecting unequal coverage, political influence and organizing resources.
Communities with media-recorded opposition averaged about 79% white, compared with 53% without recorded opposition. This measures documented resistance, not consent or access to hearings. Source: Capital B.
The next test is who can shape approval
The researchers' reported recommendations include data center-specific zoning rules, racial impact statements before construction and explicit consideration of racial equity in development decisions. These are recommendations, not protections the study itself puts into effect.
The evidence needed next is more specific than another national count: which projects receive hearings, what obligations developers accept, and whether promised jobs and revenues reach the communities hosting the facilities. Pollution and household costs also require direct measurement rather than inference from a census tract's racial composition.
For residents such as Mabery, the immediate question is whether those decisions will be open to challenge before approval. Augusta's exemption for an already authorized project shows why changing future rules and giving existing neighbors a meaningful say are separate tasks.
At Ignite, Nvidia and Microsoft are expanding local coding AI, cloud desktop streaming and Azure infrastructure. The announcements build on an existing Windows AI strategy, with AMD alternatives, staggered availability and no comparable deployment-cost results in the announcements.
Microsoft and Nvidia are expanding Azure’s Vera Rubin virtual-machine preview, with systems already running in its data centers. Microsoft’s separate confidential-AI showcase covers four GPUs, while the announcements leave pricing, regions and general availability unspecified.
Meta is reconsidering its 2030 climate deadline after net emissions reached 11.8 million tonnes in 2025. Hardware and infrastructure purchases dominate its reported footprint, while new gas generation adds a longer-term challenge.
Three House committee chairs want the Remote Access Security Act in the annual defense bill as scrutiny of Nvidia intensifies. The available Senate text would authorize controls on specified foreign users’ remote access, with computing limits left to Commerce.
Google Quantum AI reports tests of quantum contextuality spanning up to 105 qubits on Willow processors. The experiments expose hardware errors and beat restricted classical strategies, while the largest task relies on an unproven classical benchmark.
A bipartisan House disclosure shows requests to sell 3.4 million AI chips to China and approvals for more than 879,000. The oversight question is whether Commerce verified the buyers, supply protections and access restrictions behind those authorizations.
OpenAI says its internal Performance Factory tracked more than 1,700 investigations in just over three months. About 400 cases reached proposed fixes, but far fewer shipped repairs used Codex’s code unchanged, and engineers retain approval.
Activists protested Anthropic’s stock-market debut outside Nasdaq, with at least one arrest and conflicting accounts of the total. Their campaign targets retirement-fund exposure to AI shares, alongside concerns about jobs, data centers and public accountability.
Anthropic says cash, growing revenue and multiyear payment schedules support its vast compute commitments. Its filing shows why flexibility matters: short-term capacity costs more, while major cloud contracts require payment regardless of usage.
OpenAI is rolling out live transcription for dictated ChatGPT prompts across free and paid plans. Users can see words as they speak and review the text before sending, but the company has not supplied performance measurements or a replacement recording maximum.
Boltz has raised a Bessemer-led $110 million Series A to expand molecular-design tools sold by usage. Its open-source research brought broad adoption; the commercial test is whether paid computation delivers repeatable gains in laboratory discovery.
Donald Trump said he and Xi Jinping supported minimal AI guardrails but had not discussed definitions. Their Shenzhen meeting leaves unresolved how bilateral safety cooperation would coexist with U.S. opposition to global regulation.
Gemini can now click and type across Mac apps and websites in a background window, Google says. The paid rollout expands its desktop automation, but app permissions expose active accounts and file changes that backups cannot always undo.
The House AI Safety and Innovation Caucus promises legislative advocacy and rapid incident briefings. Its members already have bipartisan safety proposals in play, making the choice of safeguards its next test.
Binance Wallet says it will return all subscription commitments, distribute roughly 2,957 Anthropic tokenized shares free and add a $1 million token giveaway after demand far exceeded its allocation.
Early Anthropic investor Dustin Moskovitz is urging frontier AI developers to slow down. He and Cari Tuna moved their stake to a nonprofit in early 2025, directing potential returns toward philanthropy while leaving questions about the holding's value and governance rights.
A debut-day estimate values the Anthropic shares FTX sold in 2024 at $49.7 billion. The comparison exposes the scale of subsequent paper gains, but contrasts cash raised for bankruptcy repayments with a hypothetical stock-market valuation.
Anthropic raised $81.3 billion in a record IPO and opened 34% above its offer price before retreating. Rapid revenue growth and adjusted operating profits sit alongside vast computing commitments and limited investor control.
The Kimi developer reportedly abandoned planned Nvidia B300 capacity for 2027 after new US restrictions. The deal exposes financing risks for overseas AI clouds, but neither a model delay nor a $150 million cash loss is established.