A nationwide pattern behind the AI infrastructure buildout

As demand for cloud computing and artificial intelligence drives a new wave of data-center construction across the United States, a new study raises a consequential question about where that infrastructure is being built. Researchers at George Washington University and the Center for Security, Innovation & New Technology at American University found that data centers are disproportionately located in communities of color.

The study, titled Redlining in the Age of AI: The Relationship between Data Center Locations and Race in the US, examined 81,000 US census tracts and location data for 5,291 data centers. Its findings point to disparities that remain after accounting for several commonly cited factors in facility siting, including population, household income, education, land costs, proximity to major transmission lines, and state-level effects.

Majority-Black neighborhoods had 2.2 times the odds of hosting a data center compared with a white neighborhood in the same state, according to the researchers. The odds were 2.6 times higher for majority-Hispanic neighborhoods and 4.2 times higher for majority-Asian neighborhoods.

Infrastructure decisions have social consequences

Data centers are central to the computing systems behind consumer internet services, enterprise software, cloud platforms, and AI. Their expansion also brings substantial needs for land, electricity, grid connections, and in some cases water. Those requirements make their locations a public-policy issue, particularly where communities have limited influence over land-use decisions or must carry the effects of nearby industrial infrastructure.

The researchers reported that the observed racial disparities persisted even when the analysis incorporated variables that can influence where a developer builds. The finding matters because it suggests the pattern cannot be reduced simply to data centers clustering around large cities, which are often more racially diverse, or to the availability of low-cost land.

CSINT executive director Samantha Bradshaw told 404 Media that the researchers wanted to test whether local accounts of data-center development fit a broader national pattern. Their analysis, she said, found that neighborhoods hosting data centers tend to have a higher proportion of residents of color.

The legacy of redlining and the power grid

The study connects the present-day geography of computing infrastructure to older patterns of investment and exclusion. Historically, redlining and related discriminatory policies restricted housing finance and other opportunities for Americans of color, helping shape where people lived and where infrastructure was built.

Black Neighborhoods Are Twice as Likely To House a Data Center, Study Finds
Search media contentGas turbines on the south side of the Grok AI Data Center on Riverport Road in Memphis, Tennessee, USA, on July 31, 2026. xAI has multiple sites in the works in and around south Memphis. (Photo by Carlos Kosienski/Sipa USA)(Sipa via AP Images)

That history can intersect with a developer's practical search for reliable electricity. Data centers are often placed near transmission lines because those connections can ease access to power. But the transmission infrastructure itself was frequently developed through neighborhoods with large Black populations, according to the reporting on the study. In that sense, a seemingly technical siting decision can follow a map created by much older inequities.

Co-author Dhanaraj Thakur told 404 Media that he did not believe developers were necessarily seeking out communities of color intentionally. Rather, he described race as deeply embedded in the factors that shape development decisions. The distinction is important: disparate outcomes can emerge through systems of land use, energy infrastructure, and investment even without a single explicit discriminatory decision by a developer.

What the research does and does not establish

The study identifies an association between data-center locations and neighborhood racial composition. It does not, on the information provided, establish that every facility causes a specific local harm or that each developer made its decision for the same reason. It also distinguishes among types of data centers, noting that hyperscale facilities can differ substantially from other facilities.

Still, the scale of the dataset gives the findings weight in a debate that is often conducted one project at a time. Community disputes over individual facilities can center on electricity demand, water use, noise, diesel backup generation, land conversion, or pressure on local infrastructure. A national analysis can reveal whether those site-level controversies accumulate unevenly across the country.

A growing test for AI policy

AI policy is often framed around models, chips, jobs, safety, and competition. This research redirects attention to the physical footprint of the industry. The expansion of computing capacity depends on choices made by utilities, local governments, developers, and communities long before an AI service reaches a user.

For policymakers and companies, the study creates a clear need for more transparent siting processes and more rigorous assessment of cumulative local impacts. It also argues for treating data centers as more than neutral nodes in a digital network. They are large pieces of energy and land infrastructure, and their placement can reinforce the inequalities already present in the built environment.

The AI boom will continue to require new computing facilities. The question raised by this research is whether the benefits and burdens of that buildout will be distributed fairly.

This article is based on reporting by 404 Media. Read the original article.

Originally published on 404media.co