---
title: "Majority-Asian US neighborhoods have 4.2 times the odds of hosting a data center"
description: "A study of 5,291 facilities finds elevated odds in communities of color after adjusting for income, land and power lines"
author: "Luis Goa"
published: 2026-11-19T15:00:00Z
modified: 2026-11-19T16:10:44Z
url: https://rews.cc/a/majority-asian-us-neighborhoods-have-4-2-times-the-odds-of-h-f86d01
language: en
tags: ["data-centers", "redlining", "infrastructure", "inequality", "ai", "tech"]
publisher: "Rews (https://rews.cc)"
---

# Majority-Asian US neighborhoods have 4.2 times the odds of hosting a data center

*A study of 5,291 facilities finds elevated odds in communities of color after adjusting for income, land and power lines*

By Luis Goa · November 19, 2026 · https://rews.cc/a/majority-asian-us-neighborhoods-have-4-2-times-the-odds-of-h-f86d01

## In brief

- Odds ratios versus white neighborhoods are 2.2 for majority-Black, 2.6 for majority-Hispanic and 4.2 for majority-Asian areas
- Researchers analysed 81,000 census tracts and 5,291 data centers with economic, infrastructure and state-level controls
- Historical redlining and power-line placement may embed race in modern siting without deliberate developer targeting
- Meta’s $50 billion Louisiana project is paired with promises of 1,000 permanent jobs and a 2,000% sales-tax increase
- Recommendations include data-center zoning rules and racial-impact statements before construction

Majority-Asian neighborhoods in the US have 4.2 times the odds of hosting a data center as a white neighborhood in the same state, according to *Redlining in the Age of AI: The Relationship between Data Center Locations and Race in the US*. The corresponding odds ratios are 2.6 for majority-Hispanic neighborhoods and 2.2 for majority-Black neighborhoods, [Matthew Gault reported for 404 Media](https://404media.co/redlining-ai-data-centers-and-race-study/). The study was produced by researchers at George Washington University and the Center for Security, Innovation & New Technology at American University.

Relative odds of hosting a data center White neighborhood in the same state: 1.0× baseline Majority-Black neighborhood: 2.2× Majority-Hispanic neighborhood: 2.6× Majority-Asian neighborhood: 4.2×

Those figures are odds ratios, not the percentages of neighborhoods with facilities and not ratios of facility counts. Odds are the probability of hosting a site divided by the probability of not hosting one. A 2.2 odds ratio therefore does not mean that 2.2 times as many majority-Black neighborhoods have data centers. The researchers’ broader result is that the odds rise as the shares of Black, Hispanic and Asian residents increase.

![A massive industrial building topped with rows of cooling units looms over a neighborhood of small houses and trees.](https://rews.cc/img/947cc1d31937d6895ad0c91a54af10be9e22aae9.webp)

## 81,000 tracts and 5,291 facilities

The national analysis covers 81,000 US census tracts and location records for 5,291 data centers. It adjusts for population, household income, education, land costs and proximity to major transmission lines. State fixed effects add another control: comparisons are made within states rather than letting siting differences between, for example, two state regulatory regimes drive the national result. The racial disparities persist after those adjustments and despite data centers concentrating around metropolitan areas, which often have more racially diverse populations.

That makes the finding an association under a statistical model, not proof of intentional discrimination by developers. It shows that the included variables do not explain away the pattern. Co-authors Samantha Bradshaw and Dhanaraj Thakur said they undertook the work after reports of data-center pollution in Black and Brown communities in Memphis, Tennessee. Thakur said he doubted developers were deliberately seeking communities of color: “It’s just how embedded race is with all these other factors.”

## Power infrastructure carries historical siting decisions

The proposed mechanism begins with redlining, the denial of loans and other opportunities that confined Americans of color to particular neighborhoods. Data-center developers seek ready access to electrical infrastructure, so transmission corridors are attractive sites. Those corridors were themselves often built through majority-Black neighborhoods. A location rule that appears racially neutral—build near available power—can consequently reproduce an older geographic pattern. Transmission proximity cannot be the whole statistical explanation here, however, because it is already one of the study’s controls.

![Power lines cross a stylized neighborhood and connect to a server cabinet.](https://rews.cc/img/6ef7a5de193ec291c9e7f91eeaccf0cbc41135c2.webp)

A land-use decision in Taylor, Texas, illustrates the local version. A white family donated land to the city for a public park beside a Black neighborhood. The park was never built, and the tract is now slated for a data center. Taylor and Memphis supply mechanisms and cases to investigate; neither case contributes a separate national effect estimate.

## Hyperscalers and co-location sites are different assets

The study distinguishes large hyperscale facilities built for companies such as Meta, Google and Grok from smaller co-location facilities, which can combine network interchange functions with rented computing space. Hyperscale projects sometimes arrive with promises of local investment and employment, but the study says those projects typically are not built in communities of color. Most promising substantial economic benefits are instead in white rural communities. Combining the two facility types in one location dataset can therefore measure one racial pattern while masking differences in investment and scale.

Meta’s announced $50 billion facility in rural Louisiana shows why the benefit claims need their own arithmetic. Community leaders said it would support 1,000 permanent jobs and reported that sales-tax revenue had jumped 2,000%, or to 21 times its starting level, followed by a $50,000 bonus for area teachers. No starting dollar amount accompanies that percentage, so the revenue gain’s fiscal scale is unknown. Dividing announced investment by promised permanent employment gives $50 million per job, but that is not a measure of wages, public return or net economic benefit.

![A giant tied sack pours a few coins from a pipe onto an empty office chair.](https://rews.cc/img/440e454dbbc943a56fddf46dbf81e8f069a63582.webp)

## The proposed remedy is scrutiny before construction

The researchers recommend data-center-specific zoning rules, explicit consideration of racial equity in siting decisions and racial-impact statements before projects are approved. Such statements would make demographics an input to the permit process rather than a finding after facilities are operating. Local governments are already using blunter instruments: at least seven Bay Area cities had [paused data-center construction before the midterms](https://rews.cc/a/the-bay-area-builds-the-ai-boom-and-bans-its-buildings-030e1f). A moratorium changes whether building occurs; it does not by itself redistribute the facilities’ costs or benefits.

Bradshaw’s institutional criticism is that the people designing technology policy and technical systems do not represent their users. She described the prevailing designer as “a very specific kind of white male Silicon Valley California ideology engineer”. Gault also situates the siting disputes in a recent electoral rejection of Big Tech’s power; data centers make that power visible as land use, grid connections and local pollution rather than only as software policy.

The next useful measurement would distinguish new hyperscale capacity from co-location sites and report it in megawatts alongside race, local pollution exposure and electricity costs. Tract-level odds establish where facilities tend to appear; they do not yet show who receives how much of the physical burden or the promised income.
