The AI industry has spent two years treating its physical expansion as a question of capital and supply chains. On 3 November, 41 American communities were asked directly whether they wanted a data centre. Forty said no.
What Happened
Forty-one local ballot measures concerning data centres went before voters across ten US states on election day, by Ballotpedia's count the largest such wave recorded. By the following day, with most results in, voters had sided with the data centre position exactly once.
Ohio accounted for 21 of the measures on its own. Six comparable measures had already been voted on earlier in 2026, and three more are still scheduled.
Results worth naming
- Wells, Maine — more than 80 per cent of voters approved a retroactive 180-day moratorium on data centre development. Retroactive is the significant word: it reaches projects already in motion.
- Emporia, Kansas — voters approved a prohibition.
- Wilmington, Ohio — around 80 per cent rejected rezoning across four separate referendums tied to a data centre project.
- Harwood, North Dakota — the single exception. Voters rejected, by roughly three to one, a proposal requiring extra review of industrial projects including data centres. A $3 billion facility, Polaris Forge 2, is already under construction nearby.
The Harwood result is instructive rather than anomalous. It is a community that had already accepted a large project, voting not to add friction to it — which is a different question from whether to allow one in the first place.
What the Results Do Not Tell Us
It is tempting to read a motive into this, and the reporting on the results does not support one. These votes recorded a decision, not a reason.
The concerns that dominate local data centre debates are well known — electricity demand and its effect on household bills, water consumption for cooling, noise from cooling plant running continuously, and the fact that a facility occupying a very large site and a very large share of local power typically employs a modest number of people once built. Those arguments were certainly made. Whether they drove the votes, and in what proportion, is not established by the outcomes.
What is documented is that data centre regulation was a live campaign issue beyond the ballot measures themselves. Incoming Democratic governors Gina Hinojosa in Texas, Keisha Lance Bottoms in Georgia and Amy Acton in Ohio all made it a feature of their campaigns — and those are three of the largest data centre markets in the country.
Analysts also disagree about significance. One view holds that a set of local land-use votes is weak evidence of any national mandate on AI. Another treats the midterms as the first election in which AI infrastructure was a genuine voting issue. Both readings are defensible on this evidence.
Why This Is a Harder Constraint Than It Looks
The AI build-out has run into several limits already — chip supply, grid interconnection queues, transformer lead times, cooling water. Each of those is an engineering or procurement problem, and the industry is good at those.
Planning permission is different. It cannot be accelerated with capital, it does not respond to a better supply chain, and it is decided by people who do not benefit much from the outcome. A data centre consumes land, electricity and water in a specific place while producing a service delivered everywhere — the costs are local and legible, the benefits diffuse. That asymmetry is exactly what local democracy is built to act on, and our explainer on how data centres work and why they use so much power covers the physical demands that make it contentious.
Nor does it stop construction nationally. These were individual towns and counties; the overwhelming majority of US capacity is untouched. What changes is the risk premium: siting now carries a real chance of refusal, which lengthens timelines, raises the cost of securing land and power, and concentrates projects in jurisdictions that actively court them. For an industry whose financial case rests on building very fast, delay is the cost.
Power is the connected question — demand of this scale has to come from somewhere, and our piece on grid-scale storage and energy geopolitics covers how the supply side is responding.
The Market Noticed Within Two Days
This was not treated as a local planning story. Two days after the vote, chip and AI-linked shares fell sharply, with investors citing expected regulatory friction and slower data centre approvals among the reasons — we covered that session in the November AI selloff.
Whether that reaction was proportionate is a separate question. But it establishes something: the market now treats local planning outcomes as material to the AI trade, which is a change from a year ago.
Key Takeaways
- 41 local data-centre measures across 10 states; 40 went against development.
- Ohio alone had 21 of them; Wells, Maine passed a retroactive moratorium with over 80 per cent.
- The one exception, Harwood, North Dakota, already hosts a project under construction.
- The votes record a decision, not a motive — be careful attributing reasons.
- Planning cannot be solved with capital, which makes it a harder constraint than chips or power.
Written by the WritePures newsroom. Spotted something that needs updating? Tell our editors.
0 Comments
Join the conversation
Replying toNo comments yet. Be the first to share your thoughts.