AI Data Centers Are Now a Local Ballot Question

The remote machine on the local ballot

In several Ohio cities, voters chose on November 3 to ban data centers above 25 megawatts. El Dorado, Kansas, approved a ten-year ban on facilities exceeding a 10-megawatt peak electrical load. These are oddly specific numbers for a technology described with airy words such as intelligence, transformation and cloud. The cloud, it turns out, arrives with a power rating—and occasionally a planning application. ^[1]

Reporting published on November 4 found that voters had faced 45 local ballot measures concerning data centers, with 33 results set to restrict development as of that morning. Candidates supporting moratoriums or tighter conditions also won several prominent races. The demonstrated change is plain: resistance to infrastructure supporting AI has become an organised electoral force, although the results were not uniform and cannot be treated as a single national verdict. ^[1]

What interests me is the distance folded into the word “remote.” The computation may be remote from the person using an AI service, but the building is not remote from the person living beside it. One person receives a quick answer. Another receives construction traffic, new power and water demand, or an exhaust stack in the kitchen-window view. Convenience travels. Consequences settle. ^{[1] [2]}

That does not make every data center harmful or every opponent correct. It does make siting a question of agency. Who can alter the plan? Who may only comment on it? Who receives jobs or public revenue, and who absorbs noise, pollution risk or infrastructure costs? A public meeting can produce many voices while leaving the decision perfectly untouched. A voice is useful. A lever moves something. ^{[1] [2]}

Before input becomes aftermath

Timing matters because construction can create its own argument for continuation. A WIRED and Floodlight investigation reported that a resident beside the Stargate development in Abilene, Texas, learned what was being built only after work began in 2024. The project initially obtained minor air permits that required no public notice or comment, before its developers later sought a major permit for a much larger collection of gas turbines and generators. ^[2]

The developer, Crusoe, said the Abilene project contributed to economic development and funded fire trucks, school expansion and road improvements. That is a serious counterargument, not a corporate footnote to brush aside. Large facilities can bring investment and public resources. Communities may reasonably decide that the benefits justify the costs, particularly where employment and infrastructure funding are scarce. ^[2]

But consent becomes strange when benefits are presented after physical commitment has narrowed the alternatives. “Have your say” can quietly become “say what you like.” The semantic reversal matters because consultation is not defined by the presence of a microphone. It is defined by whether speaking can still change the scale, location, operating conditions—or existence—of the project. ^[2]

Former Texas environmental regulators interviewed in the investigation argued that meaningful participation must occur before permits are issued and momentum becomes difficult to stop. The state regulator disputed claims that industry growth had weakened enforcement, saying it remained committed to protecting public health. That disagreement is precisely why enforceable procedures matter: residents should not have to choose which assurance sounds friendlier, then discover years later which one was true. ^[2]

A voice, a lever, a choice

There is a smaller, unexpected connection. Some AI work can happen on personal devices rather than distant infrastructure. Hugging Face’s documentation for Gradium Phonon describes compact text-to-speech models—software that turns written words into spoken audio—designed to run on-device. Moving this modest speech task locally does not materially replace the vast training and inference workloads driving large campus proposals. It does show, however, that centralisation is sometimes a design decision rather than a law of nature. ^[3]

I think that distinction should enter ordinary purchasing and policy decisions. When an AI feature requires distant computation, companies should explain why. When local processing is genuinely practical, customers and institutions can ask whether it offers better privacy, resilience or control. Local operation has limits: Phonon’s documentation, for example, warns that numbers, amounts and addresses need additional text handling. Small can be useful without pretending to be mighty. ^[3]

The strongest case against local restrictions is that fragmented rules could delay useful services, raise costs and push development toward places with weaker protections. That risk is real. A town-by-town patchwork is not an elegant national infrastructure policy. Yet speed without legitimacy carries its own delays: moratoriums, political backlash and projects that communities experience as something done to them rather than built with them. ^{[1] [2]}

For anyone facing a proposal nearby, the useful moment is early and unglamorous. Watch planning agendas, utility proceedings and environmental notices. Ask what residents may actually change, who pays for grid or water upgrades, what measurements will be published, and what remedy follows if promised limits are exceeded. The November 3 votes do not prove that data centers have been rejected everywhere. They show that “somewhere else” now has voters. The remote machine has become a local decision. ^{[1] [2]}