The human cost of data centers

ChatGPT, Claude, and Gemini all seem like magic. But who are the people and rare minerals behind it?

After walking by a beautiful and intricate spiderweb the other night, I found myself wondering how long it takes spiders to weave their webs. I don’t normally use ChatGPT—but I asked it my question, and it felt like it was conjuring my answer out of thin air. I noticed myself quietly enjoying the instant gratification of receiving this spiderweb literature (before realizing that I should probably fact-check it before sharing with anyone).

Instant gratification aside, I am rationally aware that this is not magic—that the manual labor and physical resources required to generate the answer to my spiderweb question is immense. I know that AI is drawing upon hard-earned knowledge from trained human spider biologists, possibly without their consent—and also that a lot of energy output is needed to assemble the expertise to generate my answer.

But in recent months, critical and detailed reporting has emerged to ground us in the reality of the vast amount of natural resources, construction, and labor that are required for AI to produce answers to any question in the world. Whether big tech companies are extracting copper in Chilean indigenous communities or running water supplies dry in Georgia, the “ease” with which we get our AI-generated answers is far from what the truth on the ground entails.

For example today, gardeners, stargazers, and birdwatchers in Bessemer, Alabama are bracing for the build of a 4.5-million-square-foot data processing center, which would cut down 100 acres of forest. The proposed construction involves building eighteen buildings on 700 acres of wooded land currently zoned for agricultural use. Each building is larger than the average Walmart Super Center, and all of them exist to house data servers.

Many of the residents’ specific questions about the $14.5 billion project have gone unanswered, in part because the mayor and others briefed on the concrete plans have signed non-disclosure agreements with the datacenter owner. But according to recent local reporting, the data center campus “could feasibly consume around 10.5 million megawatt hours per year, which is “more than 90 times the amount of energy used by all residences in Bessemer and more than 10 times the amount of energy used by all residences in Birmingham annually.”

The details are murky still on who is behind the Bessemer center—with attorneys for the residents positing that it could even be a Chinese data center. But it could also be Mark Zuckerberg, who announced in July that Meta will spend hundreds of billions of dollars to build data centers called “Prometheus” and “Hyperion.” Just one data center is expected to amass the size of Manhattan.

I spoke with a person who previously worked at a Meta data center (and who chooses to remain anonymous for fear of retaliation). Over email, she told me that residents living near the center where she worked made numerous complaints about light pollution affecting bird migration. She further shared that the data center’s use of water and electricity affected the local town’s ability to leverage their home’s utilities, especially during summer months when temperatures require more water and electricity for cooling racks. At one point, she said, this dynamic left the town without water.

This mirrors what is happening in Bessemer, where the residents are worried about soaring power bills they can’t afford—as well as limited water supply because of Alabama’s lack of a comprehensive water plan and the gallons of chilled water that may be used to cool a hyper scale data center, dissipating the heat into the air through a cooling tower.

Nearby in Georgia, a New York Times report tells the story of Newton County, Georgia, where residents may have ration water because of the amount of water required to cool Meta’s AI calculations within local data centers. Meta’s data center “uses about 10 percent of the county’s total water use daily” according to the executive director of the Newton County Water and Sewerage Authority—and companies applying to build new centers are asking for as much as six million gallons of water a day, which is more than the county’s entire daily use. The taps of several Newton County residents have gone dry, and one person interviewed said that she was afraid to cook with the tap water because of the sediment.

Big tech of course argues that building AI is necessary to keeping America competitive and fundamental future of humanity, no matter the environmental wreckage. This is a familiar talking point adopted from big business and parroted most recently by cryptocurrency enthusiasts during the crypto “boom.” They argued that the energy costs needed for crypto’s computational calculations were part of a necessary “innovation” that will usher in a new form of “financial freedom.” (That assertion has yet to bear out.)

While predictive AI may be all the rage, history might be the best place to look here to predict the future. In her book Empire of AI, journalist Karen Hao writes about the indigenous people in Chile whose land was discovered as the largest supply of copper. The land became a major extraction site for multinational companies, but in order to process the copper, the mining also drained the region of water—at one point depleting a basin in a salt flat and eradicating the population of flamingos who were spiritual siblings of the local people.

Future generations reportedly fell into depression, alcoholism, delinquency, and lacked enough food, running water, proper health care, or educational resources. Hao writes that today, the Chilean people’s problems are not with copper or lithium mining itself; their ancestors mined too. Their concern is with the scale and speed at which it is all happening. Indigenous populations believe that their unique desert land, which is home to many microbial communities, could be potentially useful for medicines or new sources of energy that don’t exist anywhere else.

As we start to have more holistic glimpses into what is transpiring in Alabama and Georgia, the Chilean history is critical to understand. It’s a history that the tech enthusiasts don’t seem to care to acknowledge, as they blindly march towards building someone’s “better future.”

Perhaps this is why one town in Virginia is fighting the construction of an AI data center in order to preserve the land where it conducts Civil War reenactments. Maybe, as silly as it may seem to have actors running around with their uniforms and rifles, they recognize that past is prologue.

Do you have a story about working at a data center or your life near a data center? We’d love to hear about it—my Signal is @ari.steinhorn17.