Partner with Us →
Nov 6, 2026

American Voters Have Turned Against the Hyperscale Campus

On November 3, voters in 41 local ballot measures sided against data center development in all but one. The next phase of AI infrastructure will pair firm power that already exists with smaller deployments behind the meter.

Category
Perspective
Author
Jason Brink, Co-Founder & Chief Strategy & Systems Officer
← Back to articles

American Voters Have Turned Against the Hyperscale Campus

On November 3, Americans voted directly on data centers in 41 local ballot measures across ten states. By the following afternoon, the side opposing data center development had won every measure that had reported except one, in Harwood, North Dakota. The national voter survey shared this year by the major networks and the Associated Press found that 76 percent of voters would oppose the construction of a new data center in their community.

The opposition crosses party lines: majorities of every party group opposed a local data center in Gallup's polling this year, and in primaries in both parties, officials who had backed data center projects lost to challengers, defeats that local observers and one of the losers tied to those votes. We have argued in public for more than a year that compute should be built where the power already is, and in October we wrote that the large campus, built on new land and waiting on a new grid connection, was running short of public permission. The election made that case for us, and I think the next phase of AI infrastructure will be built as smaller deployments placed behind the meter at power plants that already run.

What the ballots said

Ohio accounted for 21 of the 41 measures. In Granville about 82 percent voted for a full ban. In Trenton about 80 percent voted to keep new data centers below 25 MW, a limit that blocks the annexation of roughly 600 acres for a planned 1,000 MW Amazon campus, although that project could still go ahead on township land. In Piqua a charter amendment against data centers passed with 67 percent, and more than 70 percent voted to recall three city commissioners who had supported a data center project. In Wells, Maine, more than 80 percent approved a retroactive 180-day moratorium.

The same pattern showed up in races for local office. In Pine Island, Minnesota, where Google has proposed a data center, a first-time candidate who opposed the project won about two thirds of the vote against the interim mayor. In Hermantown, the site of a second Google project, two newcomers who ran against it outpolled an incumbent who had voted to approve it.

There was one exception. Voters in Harwood rejected a proposal for extra review of industrial projects by about three to one, in a town where a US$3 billion data center is already under construction nearby. All of the November results are unofficial and will not be certified for some weeks, and outlets counting at different times on November 4 reported slightly different tallies. The direction of the results is clear in every count we have seen.

What the voter survey said

The networks and the AP pooled a single voter survey this year, run by SSRS, with more than 75,000 respondents. Asked whether they would favor or oppose a new data center in their community, 24 percent said favor and 76 percent said oppose. Fifty-six percent said AI and data centers were a crucial or major factor in their vote for Congress, and 86 percent said they were concerned about the cost of utilities. Asked which party would better handle AI, 43 percent said neither.

Polls through the year point the same way. Gallup found in March that 71 percent of Americans opposed an AI data center in their area, with majorities opposed in every party group. Pew found this summer that half of all adults now say data centers are mostly bad for home energy costs, up from 38 percent in January. In October, Heatmap and Embold Research asked registered voters which arguments against a local data center they found most convincing, and higher electricity bills came first, picked by 40 percent.

Both parties heard the same message

In Ohio, both major-party nominees for governor proposed pausing new data centers. The Republican nominee pledged an executive order on his first day halting approval of new projects, and the Democratic nominee backed a conditional moratorium on construction until developers met a list of conditions, including covering the full cost of the utility service they need. In Texas, the Republican governor began pausing data center approvals at the state grid in August, and later at the environmental permitting agency, pending an audit, and his Democratic challenger called for a moratorium on all approvals.

Some of the sharpest verdicts came inside a single party. In Box Elder County, Utah, two Republican commissioners lost their primaries after voting in May to let a state authority create the project area for a proposed hyperscale campus, and one of them said after his defeat that he believed the data center vote cost him the election. In Calvert County, Maryland, three Republican commissioners who had voted against a data center moratorium lost their primaries, and in Frederick County the Democratic council president most associated with a data center zoning expansion finished third in his party's primary. Local observers linked both results to those votes.

Statewide races turned on a great deal more than data centers, and I would not read any single governor's or Senate result as a verdict on them. The local measures are cleaner evidence, because the data center was the only question on the ballot.

What voters are refusing

As I read it, the opposition falls hardest on a particular kind of project: very large sites, on large parcels of land, whose power costs are seen to land on household bills. Not every vote drew that line, and Granville, Piqua and Wells restricted data centers of any size. In PJM, the largest grid operator in the United States, the independent market monitor attributes about US$6.5 billion, or 40 percent, of the 2027/28 capacity auction to data center load, and US$6.3 billion of the 2028/29 auction. In June the monitor told PJM's members that data centers had so far imposed US$23 billion in capacity costs on existing customers.

Local opposition was already blocking or delaying projects before anyone voted. Data Center Watch counted at least 75 projects valued at about US$130 billion blocked or delayed by local opposition in the first quarter of 2026, and at least 45 valued at about US$68 billion in the second. In February the commissioners of Montour County, Pennsylvania unanimously refused to rezone more than 800 acres of farmland owned by Talen Energy, a change that would have allowed data centers and an expansion of Talen's coal and gas-fired Montour power plant to supply them. My read is that having generation in the plan was not enough to carry a project that needed that much farmland.

The rules being written now draw their lines around size. Trenton's cap is 25 MW. Virginia's regulators created a separate rate class, starting in 2027, for high-load customers of 25 MW and above, designed to keep their costs from shifting onto other customers. New York's governor paused state environmental permits for data centers of 50 MW or more in July. Texas law requires new loads of 75 MW and above to accept curtailment in grid emergencies. A deployment below 25 MW sits under all four of these lines.

Large campuses are still being approved, including a roughly 900 MW campus in Coweta County, Georgia that passed 3 to 2 in April over a petition of nearly 8,000 signatures. SemiAnalysis estimates that local moratoriums and New York's executive order together are delaying about 2.3 GW of capacity, against 38 GW of new US data center IT capacity it expects to be delivered in 2027. That describes projects already in the pipeline, and I think the ballot results matter more for the sites that have not yet been chosen, because communities now have a direct vote and have shown they will use it.

Where the next capacity goes

The largest buyers have already moved part of the way. In March, Amazon, Google, Meta, Microsoft, OpenAI, Oracle and xAI signed the White House's voluntary Ratepayer Protection Pledge, committing to build, bring or buy new generation for their data centers and pay its full cost. Cleanview counts 59 US data centers, together more than a quarter of planned US data center capacity, that plan to build about 90 GW of their own generation on site. About 2 GW of it is operating. In February EPRI, NVIDIA, Prologis and InfraPartners announced a program to put 5 to 20 MW data centers at or near utility substations with spare capacity, aiming for at least five pilot sites in development by the end of this year.

On-site generation answers who pays for the power, but in my view it does less for local consent, because a new gas plant beside a new campus still needs a large parcel, a fresh set of permits and the consent of the people who live nearby. The deployments I expect communities to accept pair compute with firm power that already exists, meaning generation that runs around the clock whatever the weather, and size the compute to what that plant can supply. A site built that way does not need a new grid connection paid for by other customers or a new power plant, and it does not ask a town to rezone a thousand acres. In the United States, taking a large existing plant's output behind the meter draws its own objection, that it removes supply the grid was counting on. Our host plants in Thailand are small biomass plants, a different setting from the large US plants at the center of that debate. I expect there will be more of these sites and each will be smaller, which suits the way consent is now being given or withheld, one community at a time.

What we said before the vote

We have been making this argument in public for more than a year. In July 2025 our website already described Vailinor as putting clean energy and AI data centers together "directly at the power source." In October 2025 I wrote that a better model than the hyperscale region was smaller modular facilities "placed directly alongside renewable energy generation such as biomass, hydro, or solar," drawing power at the source instead of from the wider grid. In December 2025 I wrote that instead of concentrating compute in giant centralized regions, Vailinor deploys small, dense sites at the energy source and puts stranded renewable power to work. In February, after touring a biomass plant in Thailand, I argued that power had become the constraint on AI and that edge compute would ease the strain on the grid.

In May I wrote about Stratos, a proposed campus in Box Elder County, Utah, where residents had turned out against the project. I described hyperscale campuses as "land and resource grabs with compute attached" and wrote that the story was going to repeat. Two Box Elder commissioners who voted that month to advance Stratos lost their primaries in June. Later in May I wrote that the future of AI infrastructure would be distributed, with compute placed behind the meter at plants that generate around the clock, so that it adds no new burden to an already constrained grid. In June I argued that power now comes first, and that land is chosen for the energy that can reach it. Our two October articles on this site made the same case at greater length.

Where Vailinor stands

Vailinor is developing AI compute at operating renewable biomass power plants in Thailand, supplied behind the plant's meter. Biomass plants are built to run around the clock apart from planned maintenance, which makes their output firm power. The first phase of our portfolio in Thailand covers 157.1 MW of renewable behind-the-meter power across eleven operating host plants, most of which have been generating for over a decade. That figure describes site capacity, and the IT load we can run from it is lower once cooling and the plant's own needs are taken out. Each site is sized to the plant it sits beside, and each plant has been part of its local community for years.

We build in Thailand, and the United States is the first large market where voters have been given a direct say. The pressures behind the November 3 results are present in Southeast Asia too. In September a Thai national data center board called for 49 projects under construction and 117 awaiting approval to pause while new rules were written, and officials later said the government could not order construction to stop. Thailand's rules for private on-site power supply are still being finalized, and our sites are individually smaller than the campuses in the headlines. Our first site, with Kaona Power Supply in Ubon Ratchathani, is targeted for service in late 2027, subject to funding and customer commitments. We expect to be judged on the capacity we actually energize, and on whether the communities around our host plants still welcome us once the compute is running.

Sources

Published on vailinor.com by Jason Brink, Co-Founder & Chief Strategy & Systems Officer, Nov 6, 2026.