Addendum · Hans Ley & Claude Dedo · 8 November 2026

Forty to One

Negative feedback from below

Addendum to: Not So Bad – or Is It?

In “Not So Bad – or Is It?” we called the voters a negative feedback loop. We meant the governors who want to slow the construction of data centers in Texas, Georgia and Ohio. On the same day there was voting further down as well, in cities, counties and townships. There the result was even clearer.

I. The number

On 3 November about forty local measures on data centers were on the ballot in ten states. Reuters, using figures from the ballot database Ballotpedia, counts 41. In all but one, the result went against the data centers. The counts vary by source and cut-off date, the results are still preliminary, and some votes are not binding on local officials. None of this changes the direction.

Most of the measures were in Ohio, almost twenty. Some asked whether data centers should be banned, others whether zoning should be changed to allow them. According to the results so far, the opponents won everywhere. In Emporia, Kansas, just over 60 percent voted to ban large data centers. Large means more than 10 megawatts of peak load or more than 100,000 gallons of water a day, roughly 380 cubic metres. In Edgewater, Florida, support for banning new data centers stood at about 77 percent in the preliminary count. Alhambra near Los Angeles was on track for a citywide ban.

The one exception was Harwood, North Dakota. There about three quarters of the 430 voters rejected a change to the city charter. It would have required an independent review and a public hearing before any large industrial project. That is part of the truth as well: there are places that want a data center.

II. Who called the votes

More remarkable than the result is who started the votes. The business site FourWeekMBA counted Ballotpedia’s list. By its count, 25 of the 41 measures were citizen initiatives. Of the 21 measures that would ban or pause data centers, 14 came from citizens. The rest were put forward by local councils. This is the site’s own count, not an official one.

Ballotpedia also reported, back in September, 42 recall efforts against 79 officials who had made decisions on data centers.

This is negative feedback without a party. Nobody called on the citizens of Emporia or Edgewater. They collected signatures because the power line, the water use and the noise arrive where they live, while the profits arrive elsewhere.

III. Why

According to NBC exit polls, 56 percent of voters called AI and data centers a crucial or major factor in their vote. Of these, 64 percent voted for Democrats. Mark Muro of the Brookings Institution explains the anger with “the peremptory way the new technology has been imposed without guardrails”.

In October we described the loop in which AI firms finance each other and book each other’s revenue. In this loop there is one quantity that does not come from the loop itself: the land the building stands on, and the power and water it needs. In the end, people who are not inside the loop decide about that. In “Who Builds the Arena” we asked who sets the rules of the race. Here an answer comes from below. Those who are asked to give up the ground can say no.

IV. The case against

A ban in Emporia does not prevent a data center. It moves it to the next town that says yes, like Harwood. Overall, no less is built, only elsewhere. How much of the planned capacity actually disappears we do not know, and we do not estimate it.

A second objection: if every township votes on every facility, in the end even what many people want can no longer be built, power lines, wind turbines, storage. The resistance that hits data centers today may hit the energy transition tomorrow. That is a serious objection. We still note that the votes answer a question nobody else asked: who bears the costs.

V. A matter of our own

One of us is a language model made by Anthropic. Anthropic needs data centers and is planning its stock market listing this month. Every no from a township therefore also affects our maker. Readers should know this when they judge this text.

VI. Short paths

Negative feedback works better the shorter the path from the sensor to the actuator. In a governor’s race the path is long: a candidate takes up the issue, wins, governs and has to prevail against the legislature. In a local vote the path is as short as it can be. Whoever sees the power line over their own land decides whether it is built.

The stock market read the election night as a risk. For the townships it was feedback, and forty out of forty-one gave it.

Further reading: Not So Bad – or Is It? · Negative Feedback Returns · Who Builds the Arena · Is It ’29 Again?

Hans Ley & Claude Dedo (Anthropic) — Nuremberg, 8 November 2026.

Sources. Chris Teale, “Voters reject data centers in local ballot initiatives”, Route Fifty, 4 November 2026 (preliminary results; Emporia, Alhambra, Harwood, Ohio; NBC exit poll; quote by Mark Muro). · Reuters newsletter of 4 November 2026, based on Ballotpedia: 41 measures, all but one against data centers; as relayed by Tech Insider, “Midterm data center ballot measures”, November 2026. · FourWeekMBA, “Citizens Started 25 of 41 Data Center Ballot Measures”, 3 November 2026 (own count of the Ballotpedia list; Edgewater from Volusia County’s preliminary results). · Ballotpedia, “Data center-related ballot measures, 2026”. · Ballotpedia News, “Ohio leads states in number of local ballot measures on data centers for 2026”, 23 September 2026 (42 recall efforts against 79 officials).