"Inside the Dirty, Dystopian World of AI Data Centers" is a long-form magazine feature by Matteo Wong, an Atlantic staff writer, published online on 2026-03-13 and in the print April 2026 issue. The piece reports from Memphis, Loudoun County, Virginia, and Indiana, with reference to Louisiana, Phoenix, Atlanta, Dallas, and Texas, documenting the physical and environmental footprint of the AI data-center build-out and the distribution of its local costs.
Summary
Wong argues that the race to build AI capacity is remaking the physical landscape in ways visible from the ground, and that the costs fall disproportionately on communities that did not consent to absorb them, naming Memphis's predominantly Black neighborhoods, rural Indiana, and Texas. In his account the build-out produces tangible local externalities — air pollution from gas turbines, water consumption at scale, and transmission-line landscape change — that the industry's narrative reframes as "compute infrastructure." The feature serves as a numerical anchor for the physical-reality layer beneath the AI economy, paraphrasing figures from the IEA, Dominion Energy, Anthropic, and OpenAI.
Energy and water scale
Wong's central operational example is Colossus, the xAI facility in Memphis associated with Elon Musk. He reports that Colossus uses electricity equivalent to roughly 200,000 American homes annually and consumed about 11 million gallons of water in September alone, which he compares to the yearly use of about 150 homes. Per Musk, Colossus together with two nearby facilities will require roughly 2 gigawatts in total, which Wong renders as about twice Seattle's electricity demand. Wong describes a Block AI cabinet as using "dozens of times" more electricity than a traditional cabinet, cooled by water plates atop the chips.
On announced capacity, Wong cites OpenAI's stated data-center targets of more than 30 gigawatts in total, which he characterizes as "more than the largest recorded demand for all of New England." He reports that combined AI capital expenditure by Amazon, Microsoft, Meta, and Google since the launch of ChatGPT in November 2022 exceeds $600 billion, and frames this as larger, in inflation-adjusted terms, than the federal government spent building the entire interstate-highway system.
For industry-wide US data-center demand by 2030, Wong reports conservative analyses at roughly the equivalent of 40 Seattles and aggressive analyses at roughly 60 Seattles in half that time. He cites the IEA's Siddharth Singh for the projection that by 2030 US data centers will consume more electricity than all heavy industry combined (cement, steel, chemical, auto, and other industrial facilities), with roughly half attributable to generative-AI-specific facilities, and the IEA estimate that data-center emissions could more than double by 2030. As a point of global comparison, he notes that China in 2024 produced nearly as much electricity as the United States, Europe, and India combined.
Siting and local concentration
In Loudoun County, Virginia, Wong reports 199 operating data centers plus roughly 30 in the pipeline, amounting to 13% of global data-center capacity within 520 square miles, and notes that data centers fund nearly 40% of the county's budget. He identifies Buddy Rizer, Loudoun's executive director of economic development, as the operational architect of the Data Center Alley build-out, and reports 12 substations within a 1.5-mile radius of Rizer's office. He presents OpenAI's "AI Economic Zones" proposal — described in the piece as "little Loudouns everywhere" — as a policy ask that would generalize the Loudoun pattern nationally.
The Memphis reporting centers on the xAI power plant, which Wong states is built by xAI itself rather than the utility and uses up to 35 railcar-sized natural-gas turbines for power. He attributes the imagery establishing the turbine count to the Southern Environmental Law Center. KeShaun Pearson of Memphis Community Against Pollution drove Wong through the predominantly Black southwest Memphis neighborhood next to the facility; Wong reports that both he and Pearson experienced physical symptoms (cough, throat scratch) from the air. Wong documents cooling methods across facility types: industrial fans and rooftop cooling towers for traditional centers, water plates atop chips for AI cabinets, and "indoor rain" when cooling fails.
For Virginia's grid, Wong cites Dominion Energy forecasting 5.5% annual electricity-demand growth and total demand doubling by 2039, with spokesperson Aaron Ruby describing the "largest growth in power demand since the years following World War II."
Energy-mix tradeoff
Wong reports an industry tilt toward natural gas in the near term. He quotes Sam Altman, on "Conversations with Tyler," saying "Short-term: natural gas," and reports a Louisiana utility building three natural-gas plants for a single Meta data center, described as among the largest in the hemisphere, along with the extension of coal-plant lifespans. He notes that Anthropic published a report arguing the US should streamline permitting for data centers and power plants to keep pace with China. Princeton climate modeler Jenkins summarizes the industry consensus as "the market has converged on Add gas now, and then add nuclear later."
Wong frames two scenarios. The optimist's case, which he reports skeptically, holds that advanced nuclear will obviate new fossil plants by 2030 and that AI tools will invent technologies to address climate change. Wong's own framing for the pessimist's case is that "clean air may have to wait."
Externality distribution
Wong's central political claim is that the costs of the build-out fall on rural communities, Black neighborhoods, and ratepayers who did not consent to host it. This argument connects AI Data Centers to the data-center opposition that Beyond Misuse: Artificial Intelligence, Grievance, and the Future Landscape of Political Violence — Yannick Veilleux-Lepage (Combating Terrorism Center at West Point, April 2026) argues is reshaping political violence, and to the local pushback that Shroff documents in The AI Backlash Could Get Very Ugly — Lila Shroff (The Atlantic, May 13 2026) and Bassett documents in Data centers are coming for rural America — Abigail Bassett (The Verge, May 13 2026).
Provenance and caveats
The piece is a magazine feature in which Wong paraphrases figures from the IEA, Dominion Energy, Anthropic, and OpenAI; these figures are load-bearing for downstream wiki pages but warrant verification against primary sources (IEA reports, Dominion's IRP filings) when used in policy briefs. The "more than the interstate-highway system" inflation-adjusted comparison is Wong's framing, while the underlying figure of roughly $600 billion in post-ChatGPT AI capex is standard. Pearson's symptom-reporting and Wong's own throat scratch are anecdotal — useful as ground-level observation rather than epidemiology. Several of the figures carry differing confidence: the Colossus annual-electricity figure (about 200,000 American homes) and the post-ChatGPT hyperscaler capex figure (more than $600 billion) are well supported; OpenAI's "AI Economic Zones" proposal and Anthropic's permit-streamlining report are documented in published proposals; the Memphis Colossus turbine count (up to 35 natural-gas turbines) rests on SELC imagery; and the IEA projection that by 2030 US data centers consume more electricity than all heavy industry combined depends on the assumed growth rate.
Relationships
- supports: AI Data Centers — primary numerical anchor source for energy and water scale claims
- supports: AI Environmental Impact — emissions, water, air pollution
- supports: Data Center Siting / AI Power Politics — Memphis, Loudoun, Indiana, Louisiana, Phoenix examples
- supports: Nuclear PPAs for AI — "gas now, nuclear later" industry consensus
- supports: xAI — Colossus operational facts
- related: Data centers are coming for rural America — Abigail Bassett (The Verge, May 13 2026) — same physical-reality terrain, Maine-focused; complements Wong's Memphis/Loudoun reporting
- related: The AI Backlash Could Get Very Ugly — Lila Shroff (The Atlantic, May 13 2026) — Shroff's backlash piece treats Wong's data-center reporting as one of the political mobilizers
- related: Beyond Misuse: Artificial Intelligence, Grievance, and the Future Landscape of Political Violence — Yannick Veilleux-Lepage (Combating Terrorism Center at West Point, April 2026) — the infrastructure-as-target framing in Veilleux-Lepage's paper rests on facilities of the kind Wong documents
- regulated-by: Dominion Energy (if/when created) — Virginia's primary AI-power utility