Whose Lungs Pay for AI: The Hidden Cost of Data Centers

Whose Lungs Pay for AI: The Hidden Cost of Data Centers

by Anonymous · July 03, 2026

In Boxtown, a neighbourhood on the southern edge of Memphis where the streets run flat toward the Mississippi and the air carries the metallic tang of the refinery, the noise arrived before anyone knew what it was. A low, continuous mechanical drone, the sound of dozens of gas turbines spinning around the clock. By the summer of 2025, residents of this predominantly Black community could stand in their gardens and watch the heat shimmer rising off a sprawling industrial site that had appeared, almost overnight, behind a chain of fences and non-disclosure agreements. The site was Colossus, the supercomputer built to train the artificial intelligence models of Elon Musk's company xAI. To power it, the company had installed as many as thirty-five portable methane gas turbines, most of them operating without the air permits that, as one veteran environmental lawyer put it, every set of turbines he had ever encountered was required to hold. It was the opening chapter of a fight that, by the spring of 2026, would harden into a federal lawsuit.

The people of Boxtown did not ask for a data centre. They were not consulted about it in any meaningful way. They derive almost none of the economic benefit from the chatbots and image generators that the facility's tens of thousands of graphics processors were assembled to produce. What they got instead was the exhaust: an estimated two thousand tonnes a year of smog-forming nitrogen oxides, according to filings cited by the Southern Environmental Law Center, layered onto a neighbourhood that the American Lung Association had already graded an F, in a part of Memphis that the local state representative Justin Pearson describes as hosting twenty-two of the thirty largest industrial polluters in the state of Tennessee. South Memphis has child asthma hospitalisation rates among the highest in the country and cancer rates that researchers have linked to its decades of accumulated industrial emissions. The turbines were simply the newest insult in a very long sentence.

Boxtown is not an aberration. It is a preview. As the AI boom collides with the physical limits of the electricity grid and the water table, the pattern visible in South Memphis is repeating across the United States with grim consistency. The communities absorbing the noise, the diesel particulates, the groundwater draw and the rising electricity bills of AI infrastructure are, again and again, the communities with the least political power to refuse it and the least access to the technology that demand is supposedly serving. The cloud, that weightless metaphor we use for the digital economy, turns out to have a very specific postcode, and it is rarely a wealthy one.

The Arithmetic of an Unequal Burden#

Begin with the bills, because the bills are where the abstraction becomes a number on a household's kitchen table. In February 2026, the Environmental and Energy Study Institute, a non-partisan body founded by members of the United States Congress, published an analysis by its researcher Miguel Yañez-Barnuevo laying out the disparity in stark terms. Low-income residents, renters, and Black and Hispanic households in the United States can spend as much as twenty per cent of their income on energy, the institute found, against roughly three per cent for higher-income households. That is not a marginal gap. It is the difference between energy as a line item and energy as a recurring crisis, the kind that forces a choice between cooling the home and filling the fridge.

This is what researchers call energy burden, the share of household income consumed by keeping the lights on, the home warm in winter and survivable in summer. The American Council for an Energy-Efficient Economy, which has tracked the metric for years, has found that one in four low-income households spends more than fifteen per cent of its income on energy, with the figure climbing far higher in particular cities. In Baltimore, the council reported, the most burdened quarter of low-income households pay an average of around a quarter of their income on energy bills alone. As of early 2026, roughly twenty-one million American households, about one in six, were behind on their utility payments.

Now layer the AI boom on top of that. The institute noted that utilities received requests in 2025 for at least seven hundred gigawatts of new power connections, a figure that exceeds the entire electricity consumption of the United States in 2023. Data centres are the engine of that demand. They do not simply consume electricity; by competing for scarce generation and transmission capacity, they push the wholesale price of power upward for everyone connected to the same grid. The national average electricity price had climbed to nineteen cents per kilowatt-hour by the end of 2025, a twenty-seven per cent jump...