Towns were promised an AI jobs boom. They got a fence, a hum, and a water bill.

The pitch that sells an AI data center to a small town has three parts: jobs, tax money, and a shot at becoming a tech hub. On July 13, 2026, telecom reporter Karl Bode published a point-by-point case that all three are routinely false, written in response to a June essay in The Atlantic arguing the data center backlash is overblown. We build on the compute these facilities produce, so we read both sides. The checkable claims keep failing checks.

The latency claim is the easy one to check

The Atlantic's economic case leans on Michael Mandel of the Progressive Policy Institute, who argues that businesses using AI will cluster around data centers because being close to the compute shaves latency. This is our lane, so let's check it. Sitting next door to a data center instead of 300 miles away saves single digit milliseconds. Model inference takes hundreds of milliseconds to whole seconds. Nothing about using Claude or ChatGPT improves at that margin, and the rare workloads where milliseconds do matter colocate inside the facility, not in an office park near it. A town does not become a startup magnet by being 5ms closer to a GPU rack.

The jobs half of the pitch is thinner. Most of the work is construction, it is temporary, and nothing stops an operator from bringing in out-of-town subcontractors to do it. Once the building is up, a hyperscale facility runs on a small permanent crew. Quartz's May 2026 look at the employment data, which Bode cites, found tiny permanent job counts behind the headline investment figures.

Three towns, one pattern

Bode names three places to ask about water. In Cheyenne, Wyoming, officials say a 715,000 square foot Meta data center is responsible for contaminating the city water system with a rare bacterium, per Fortune's July 10 report (Bode's piece says Wisconsin; his own sources say Wyoming). In Memphis, xAI's Colossus site is facing a civil rights lawsuit brought through the Southern Environmental Law Center over pollution from the gas turbines powering it, while Politico reports the company pitching a water reuse project to ease the pressure. And in Mansfield, Georgia, the BBC interviewed residents next to Meta's facility who say their taps went bad after construction started.

The Atlantic's defense is aggregate: data centers directly consumed 66 billion liters of water in 2023, while American golf courses used nearly 2 trillion. Both numbers are real. They are also beside the point, because water harm is local. A national golf statistic does not fix anyone's well in Mansfield, and Newsweek's mapping shows much of the new buildout landing in drought-hit areas. Futurism has separately documented local power rates climbing around new facilities.

Why the buildout is courting tribal land

The tribes in Bode's headline are the newest part of the story. The New York Times reported on July 9, 2026 that developers are pitching data centers to Native American tribes because energy projects on nontribal land can wait 3 to 10 years for permits, while tribes hold sovereign authority over their own permitting and can move much faster. That is regulatory arbitrage, and the communities absorbing the risk have the least leverage to litigate against Meta or xAI if the promises fail. Bode's sharpest warning is about what happens if the AI bubble deflates the way critics like Ed Zitron expect: a glut of capacity, half-built warehouses, and towns and tribes that traded land and water for a boom that never arrived.

Why a build studio cares

Every Claude call in our builds terminates in a building like these. If the buildout overshoots demand, the glut eventually shows up in compute prices. If the backlash keeps winning town votes, siting slows down and the capacity growth every AI roadmap assumes gets harder to deliver. Either way, where the compute physically lands is part of our supply chain, and we would rather understand it than be surprised by it.

The honest caveats: Bode is a polemicist, and he picks the facts that fit the fight. The Atlantic is right that data centers' direct national water draw is small next to agriculture or golf, and these facilities do pay property taxes that some counties genuinely bank on. But the claims that sell a specific facility to a specific town, permanent jobs, a startup cluster, no impact on your bills, are the checkable ones, and those are the ones failing. Bode's suggested default toward data center promises is 'extreme skepticism and alarm.' You do not have to go that far to conclude that a town negotiating with a hyperscaler should get the jobs numbers in writing.

Next step: read Bode's full piece next to The Atlantic essay it responds to, and skim Gallup's polling. If you are building on AI compute and want an architecture that does not assume inference stays cheap forever, write to us at hello@gattyworks.com.