The 2026 midterms were about control of Congress, but a quieter fight played out on local ballots: whether to let data centers into the neighborhood. On Tuesday, November 3, voters in at least 38 jurisdictions passed bans or restrictions, according to The Washington Post, which called it “a major setback to the AI industry.” Here is what the early returns show.

The scorecard so far

According to a tally from American Buildout, voters in 10 states approved every binding data center restriction counted in local returns and rejected all 13 measures that would have advanced development. Of 38 counted measures, 36 went against development, including 21 binding restrictions and 15 bans. These results are unofficial, and a few measures were not yet confirmed or lacked official results.

Ohio leads the way

Ohio accounted for 21 of the measures, and voters approved 11 bans, including in Conneaut, Granville, Hubbard, Pataskala, Trenton and Urbana. Hubbard voted 2,622 to 802 for its ban. Granville’s charter amendment received 82.4 percent support and Pataskala’s received 73.6 percent. Six Ohio bans target facilities exceeding 25 megawatts, and ten amended city charters.

Perry, a village in Lake County, approved a requirement that developers seeking tax abatements or similar deals post financial guarantees, with 87.1 percent support, the widest margin among the binding restrictions. In Ashville, Ohio, voters rejected a resolution approving a development agreement with a data center developer.

Beyond Ohio

The pushback was not limited to one state:

  • Illinois: Bloomington backed a moratorium with 77.6 percent. In Will County, 78.9 percent of Joliet Township voters said no on whether the city council should approve AI data centers.
  • Kansas: El Dorado’s ban passed 2,966 to 1,183, and Emporia’s measure passed with 60.1 percent, though its ordinance cannot take effect until legal questions are resolved.
  • Maine: Voters in Wells, Wiscasset and Peru approved temporary moratoriums.
  • California and Florida: Alhambra approved a ban with 81.8 percent, and Edgewater with 77.2 percent.
  • Wisconsin: Janesville approved a voter-approval requirement for projects over $450 million with 62.8 percent.

Why voters are saying no

The ballots do not share a single message, but the measures target the same concerns residents raise at public hearings: electricity demand, water use, land, noise and who pays for infrastructure. Many of the measures set size thresholds, such as 25 megawatts in six Ohio bans and 10-megawatt definitions in two Kansas cities, which means they aim at large facilities rather than every server room.

What it means for the AI buildout

Tech companies are spending enormous sums on new computing capacity, and local approval is a key step. Bans and moratoriums can delay or relocate projects, and each one adds uncertainty for developers. At the same time, local votes do not stop the industry nationwide. Companies may shift to places with more welcoming rules, negotiate community benefits, or move forward in places that did not vote.

Caveats

Results are unofficial and subject to certification. Some measures face legal review, and a number of ordinances may be challenged. Boulder City, Nevada, canvasses on November 13, and some Illinois counties are still counting mail and provisional ballots through November 17, per American Buildout. Expect the numbers to change slightly.

What to watch next

  • Certified results and any court challenges to the new bans.
  • How developers respond, from community-benefit offers to site changes.
  • Whether state legislatures or regulators take up similar rules in 2027.

The bottom line

Election Day showed that for many communities, the question is no longer whether data centers are good for the economy in general, but whether they belong next door. That is a new kind of political pressure on the AI industry, and it is happening one ballot measure at a time.

Image: Data center infrastructure in the United States by DOE/National Renewable Energy Laboratory (NREL) via Wikimedia Commons, Public Domain Mark 1.0