For two years, the fight over data centers has played out one county zoning hearing at a time. On Tuesday, November 3, it moved to the ballot box — and the builders lost almost everywhere. Voters in at least 38 jurisdictions approved bans or restrictions on data center construction, according to a Washington Post tally, and Ballotpedia counted 40 of the 41 local data center measures decided that day going against development. Three of the country’s biggest data center states — Texas, Ohio, and Georgia — elected governors who campaigned on pausing or conditioning new construction. The results do not stop the AI buildout. They do change its geography, its timeline, and the price of admission.
What voters actually decided
The numbers need care, because several tallies circulated in the days after the election and they measure different things. Ballotpedia tracked 41 local data center measures across 10 states on November 3: 28 of them would have restricted development if approved, and 13 would have authorized or supported it. Voters approved 27 of the restrictive measures and defeated all 13 of the supportive ones, for a combined 40 of 41 against development. The Washington Post’s count of jurisdictions where bans or restrictions passed started at 29 on election night and rose to at least 38 as returns came in. All results remain unofficial pending certification, and the tallies differ in scope — but the direction is not in dispute.
The single exception proves the pattern. In Harwood, North Dakota, voters rejected a measure that would have required additional public review of large industrial projects, including data centers — 325 against, 105 in favor, according to the North Dakota Secretary of State. Harwood is a town of roughly 800 people, and the project at issue is Applied Digital’s Polaris Forge 2 campus, a roughly 900-acre development due to reach full operation in early 2027. Everywhere else, the restriction side won.
The measures were not uniform. Some were outright bans: ten measures banned facilities above 25 megawatts — nine in Ohio and one in Ottawa, Kansas — and two in Kansas banned facilities above 10 megawatts. Others required voter approval before a project could proceed, imposed moratoriums while a community studied the question, or were advisory votes that express opposition without blocking construction by themselves. That variety matters for developers reading the results: a ban and an advisory vote are very different obstacles, but both signal the same thing about local consent.
Ohio, the epicenter
Ohio was the center of gravity. The state had 21 of the 41 measures Ballotpedia tracked, spread across 18 municipalities, and every one went against development. The margins were not close. In Hubbard, about 76.6 percent of voters backed a prohibition on data centers within city limits. In Trenton, a ban on large data centers passed roughly 80 percent to 20 percent. In Sunbury, unofficial returns showed a ban on facilities of 25 megawatts or more passing unanimously — a measure aimed, by most accounts, at a proposed $2 billion Amazon site. All ten charter measures backed by the grassroots group Conserve Ohio passed.
The Ohio results land on top of a state that already hosts about 200 data centers, and they come with a governor-elect who campaigned on conditions rather than an outright ban. Amy Acton, the former state health director, defeated Vivek Ramaswamy on a conditional freeze: new projects could proceed only if developers cover all of their utility costs, use union labor, build on old industrial sites rather than farmland, and meet transparency and community-benefit requirements. Preliminary returns put her at roughly 51 percent to Ramaswamy’s roughly 46 percent. She has also promised clawbacks on the state’s data center incentives, which reporting puts at about $1.6 billion.
Outside Ohio, the pattern held. Nearly 80 percent of voters in Joliet Township, Illinois, opposed an advisory proposal favoring AI data centers. More than 60 percent of voters in Janesville, Wisconsin, required another public vote before major development at a former General Motors site. In Edgewater, Florida, a charter ban passed 77 percent to 23 percent, and in Maine, eight towns voted on temporary moratoriums. The opposition is not a regional phenomenon — it showed up in the Midwest, the Southeast, New England, and the Plains.
The governors’ races
The local measures were the direct votes. The governor’s races were the larger prize, and data centers were a campaign issue in all three of the biggest ones. In Texas, state Representative Gina Hinojosa defeated three-term Governor Greg Abbott after campaigning on an immediate moratorium on new data center approvals until the legislature writes protections giving Texans a say in whether and how the facilities come to their communities. She also promised stricter rules on water and power use and an end to tax breaks for the industry. Abbott himself had already directed the Texas Commission on Environmental Quality in September to pause environmental permits for data centers pending an audit — the moratorium debate in Texas was over who would control the pause, not whether one should exist.
In Georgia, former Atlanta Mayor Keisha Lance Bottoms defeated businessman Rick Jackson with about 54 percent of the vote, on a pledge of a temporary statewide pause to study the effect of data centers on power bills, water supplies, and communities. In Michigan and New Mexico, the Democratic winners also backed pauses, and the governors of New York and Oregon — where pauses were already in place — pledged to extend them. None of these pledges is a statewide ban yet. Governors’ powers vary by state, several will need their legislatures, and the promises differ in strength — an immediate moratorium in Texas, a temporary pause in Georgia, a conditional freeze in Ohio. But the electoral logic is now established. An NBC exit poll found 56 percent of voters said AI and data centers were a crucial or major factor in their congressional vote. A September Quinnipiac poll found nearly three in four Americans oppose a data center in their community, up from about two-thirds in March. The grievance underneath is the electric bill: residential power prices are up more than 10 percent nationally since 2024, and more than 20 percent in Ohio, by Reuters’ count. Most of that increase is not caused by data centers — fuel costs and decades of deferred grid investment carry most of the blame — but data centers are the visible new load on the system, and voters have connected the two.
Why 2027 is mostly safe and 2028 is not
The most important question for the industry is what the election actually delays, and the answer splits by year. Most of the capacity due to come online in 2027 is already permitted and under construction. Nomura has noted that point, and it is the reason analysts expect the disruption to land in 2028 rather than next year. A pause on new approvals does not revoke a permit already granted, and it does not stop a building already going up.
The 2028 pipeline is where the exposure sits. UBS estimated after the election that roughly 30 percent of expected US data center capacity additions in 2028 could be at risk from construction pauses, with Texas alone accounting for about 18 gigawatts of that pipeline. A separate analysis by the market intelligence firm Cleanview, cited by Bloomberg, put about 40 percent of pending US data center capacity in states whose newly elected governors campaigned on stopping or pausing construction. Bloomberg Intelligence analyst Mandeep Singh estimated the results could trim 5 to 7 percent from the roughly $1.4 trillion technology companies were expected to spend on AI infrastructure in 2027, translating to as much as 10 gigawatts of capacity failing to come online next year as planned.
The market reaction was swift. On Thursday, November 5, the S&P 500 fell more than 4 percent and the Nasdaq 100 fell 6.4 percent, erasing about $2.3 trillion in market value, as investors priced in tighter regulation of the AI buildout. The selling hit the suppliers first — chipmakers, server and networking vendors, and power equipment makers — rather than the hyperscalers themselves, whose 2027 capacity is largely secured.
The pipeline was already tightening
It would be a mistake to read the election as the cause of the slowdown. The pipeline was already constricting before a single ballot was cast. Wood Mackenzie reported on October 30 that US utilities have committed to connect 201 gigawatts of new data center capacity, and that the grid connection wait now runs 54 months at best in Virginia and 84 months in Ohio — meaning a project starting in Ohio today would not have grid power until 2033. The firm counted about 400 moratoriums on data center proposals already in place, with about a fifth of projects under development held up by at least one of them.
The election did not create the bottleneck. It converted a slow-moving permitting problem into a political one, and it did so in the states where the pipeline is largest. A moratorium is, in effect, a tax on immobile capital: a data center is hundreds of megawatts of steel bolted to a substation, with a transformer on a three-year backorder and a twenty-year lease behind it. It cannot move to a friendlier jurisdiction the way a factory or a headquarters can. When permission becomes the binding constraint, the projects most at risk are the ones that have not yet secured it — which is why the 2028 pipeline, not the 2027 one, carries the exposure. Some projects will move to neighboring states; one analyst noted that Texas projects without electricity and environmental permits lined up would “probably go to Louisiana, Oklahoma or nowhere.” But neighboring states have their own queues, their own water constraints, and — after Tuesday — their own voters watching the same issue.
How developers are adapting
The industry’s most visible response came two days after the election. On November 5, OpenAI published a National Community Compact and a companion Compute Campus Policy Blueprint, committing every future campus it owns and operates to pay the full cost of the electricity and grid infrastructure it uses, to post financial guarantees before major spending so other ratepayers are protected if a project is delayed or canceled, to end nondisclosure agreements with public officials during site negotiations, and to publish site-level water, electricity, noise, and hiring data. Each future campus would get its own local compact shaped by residents’ input, with a yearly public progress report. OpenAI said the documents had been in preparation for months and were not a response to the election — a claim that is plausible and, in a sense, beside the point. The compact reads as a direct answer to every grievance voters expressed on Tuesday: who pays for power, who knows what is being built, and what the community gets in return.
The compact has limits worth noting. It covers only campuses OpenAI owns and operates, not the partner-built sites — including those developed with Oracle — that supply much of its computing capacity. It is a set of voluntary commitments, not a contract with enforceable penalties, and it is not clear how compliance would be independently verified. But as a template it matters, because it concedes the central premise of the opposition: that developers, not ratepayers, should bear the full cost of serving new load, and that secrecy in site negotiations is no longer tenable. Other developers will face pressure to match it, and the blueprint asks state utility commissions to make large-load tariffs and public reporting mandatory for everyone — a notable case of a builder asking to be regulated.
What to watch next
Two things will determine how much of the 2028 pipeline actually slips. First, the legislative sessions. The Texas, Ohio, and Georgia governors-elect all take office in January, and their pauses will take shape through executive orders, agency appointments — including to public utility commissions — and legislation. The local measures, by contrast, take effect immediately: as one energy policy expert at Case Western Reserve University noted, cities and townships control zoning and property taxes, so the ballot measures will have a more direct and immediate effect than any gubernatorial action.
The second is the industry’s own conduct. The election was, above all, a referendum on trust: voters who believe developers will pay their own way and tell the truth about water and power use have little reason to ban them. The developers that publish their numbers, pay their grid costs, and negotiate in public will find the next siting fight easier. The ones that do not will find that the ballot box is now open for business — and that, unlike a zoning hearing, it does not adjourn.
Siting has always been where the data center boom meets democracy. Until Tuesday it met democracy one hearing at a time, in rooms where developers usually had the better lawyers and the deeper pockets. Now it meets democracy at the polls, where a majority of voters in a town can decide that a 25-megawatt building is not welcome, and no interconnection study can overrule them. The industry’s task is no longer just to find power and land. It is to earn permission — and after November 3, permission has a price, a process, and a deadline.