Voters hate AI data centers. They still do not vote on them

A new Times / Siena poll puts numbers on a fight that already shows up at zoning meetings: many voters do not want AI data centers near them. The catch is just as important. They dislike the facilities, but most do not yet treat them as a top election issue.

Quick Take

• Fact: The New York Times and Siena University released polling on Tuesday from 1,503 likely voters surveyed in early September. Asked whether they support or oppose building data centers to power AI technology, 61 percent said they oppose construction. Only 14 percent said they strongly support it.
• Why it matters: AI companies talk about compute like it is mostly a technical problem: more chips, more power, more cooling, more scale. Local politics treats compute like a land-use problem with neighbors, utilities, water systems, county boards, and bills.

  Those two realities are now colliding.

  For builders, the main lesson is not that AI data centers are doomed. The le

• Who cares: Founders and infrastructure leads should care because multi-year GPU plans depend on physical capacity that communities may resist.
• Judgment: Fairly hyped as a real public-opinion drag on AI infrastructure siting; overhyped if you treat 61 percent opposition as proof of a national ban or a finished midterm realignment. Fact: The Verge reports that Times /

What happened

The New York Times and Siena University released polling on Tuesday from 1,503 likely voters surveyed in early September. Asked whether they support or oppose building data centers to power AI technology, 61 percent said they oppose construction. Only 14 percent said they strongly support it.

The partisan cut is not a clean left versus right story. The Verge reports that 2024 Donald Trump voters were almost split, with 49 percent supporting construction and 45 percent opposing it. People who voted for Kamala Harris were much more negative, with 74 percent opposed. People who did not vote also leaned against construction, at 67 percent opposed.

The most useful detail is inside the opposition camp. Among voters who opposed AI data centers, 56 percent favored limits while 38 percent preferred a total ban. That matters. The poll does not show a pure stop-everything bloc. It shows a larger limits-first bloc that wants constraints, disclosure, and some control over where these facilities go.

The reasons for opposition are also practical, not just anti-AI vibes. The top concern was environment and water use at 32 percent. Impact on local communities came next at 21 percent. General distrust or dislike of AI was 18 percent. Regulation concerns were 10 percent, and the economy was 9 percent.

So yes, AI itself is part of the backlash. But the leading complaints are about the physical layer: water, power, noise, land use, ratepayer pressure, tax deals, and local disruption. In other words, people are not only arguing about chatbots. They are arguing about what the AI buildout does to the places where they live.

The Verge also notes that the Times / Siena result fits earlier national polling. NBC News polling in March showed AI polling poorly, and Gallup polling in May found roughly seven in ten Americans opposed local construction of AI data centers. This is not a single weird survey. It is another receipt in a growing pile.

Then comes the split-screen moment. When voters ranked the issues most important to their midterm vote, AI and data centers barely showed up. Across most demographics, the issue registered at less than 1 percent. Among voters ages 18 to 29, it reached 3 percent. Bigger issues such as the economy, foreign policy, inflation, and partisan opposition still carried more weight.

On party trust, neither side owns a clear national advantage. The Verge reports that among respondents asked which party they trusted to handle the issue, 40 percent favored Democrats, 42 percent favored Republicans, and 18 percent did not know or refused.

Why it matters

AI companies talk about compute like it is mostly a technical problem: more chips, more power, more cooling, more scale. Local politics treats compute like a land-use problem with neighbors, utilities, water systems, county boards, and bills.

Those two realities are now colliding.

For builders, the main lesson is not that AI data centers are doomed. The lesson is that capacity has politics now. A model roadmap can depend on a permit hearing. A product launch can depend on a grid interconnection queue. A cloud region can depend on whether a county thinks the trade is worth the stress.

Three reads stand out.

1. Opposition is broad, but not all opposition is absolutist.

Sixty-one percent opposition is a cold starting point for any build-more-faster message. But the limits-versus-ban split matters. A majority of opponents favored limits rather than a total ban. That leaves room for projects that can prove lower water use, clearer community benefits, credible grid planning, and real local accountability.

2. The public is reacting to operations, not slide decks.

The top concerns are water and local impact. That means the winning argument will not be a vague promise that AI will change everything. It will be concrete: how much water, where the power comes from, what happens during peak load, how loud the site is, what backup systems are used, who gets the tax benefit, and who pays if the local grid gets stressed.

3. Low salience changes the strategy.

Voters can dislike something and still not vote on it. Right now, AI data centers look less like a national ballot driver and more like a local veto weapon. That can still be powerful. A low-salience national issue can become a high-salience county fight very fast when the proposed site is down the road.

For infra teams, the clean translation is this: siting risk is product risk. If capacity is delayed, constrained, or made more expensive, that touches price, latency, availability, model development, and enterprise reliability. The cluster is not someone else's problem anymore.

Who should care

Founders and infrastructure leads should care because multi-year GPU plans depend on physical capacity that communities may resist.

Cloud buyers should care because a promised region is only useful if capacity can actually be built and connected.

Local officials should care because the trade is no longer abstract. Jobs and tax base have to be weighed against water, grid stress, traffic, noise, and public trust.

AI labs and cloud communications teams should care because national hype will not answer a resident asking about diesel backups or rate hikes.

Builders should care because the software story now rides on concrete, coolant, and public process.

If you only read AI as software, this poll looks like background noise. If you read AI as infrastructure with a better demo reel, it looks like an early warning.

What to do this week

Do not rewrite your roadmap around one poll. Do rewrite the risk section.

First, separate dislike from ballot power. High opposition plus low election salience means the pressure is more likely to show up locally than as a clean national mandate.

Second, lead with water, power, and community load when explaining capacity plans. Benchmarks and model scores do not answer the concerns that top this poll.

Third, stress-test every region you depend on. Look at grid queues, water constraints, local news coverage, tax fights, and permitting history.

Fourth, treat limits as the live lane. Since most opponents preferred limits over a total ban, serious efficiency commitments, capped draw, reuse plans, disclosure, and community benefit agreements may matter more than another trust-us campaign.

Bottom line

The public is not cheering for the physical layer of the AI boom. Times / Siena found that 61 percent of likely voters oppose AI data-center construction, with water and local impact leading the objections. Earlier national polling pointed in the same direction.

But this is not the same thing as a midterm earthquake. The issue still ranks low as a vote driver for most groups, and neither major party has a clear trust advantage on it.

The Bandwagon read: AI's capacity problem is becoming political, local, and underpriced. The hype cycle loves model demos. The buildout has to survive county meetings.

Bandwagon Check

Fairly hyped as a real public-opinion drag on AI infrastructure siting; overhyped if you treat 61 percent opposition as proof of a national ban or a finished midterm realignment. Fact: The Verge reports that Times /

Sources

• AI and data centers are incredibly unpopular in every poll
• AI and data centers are incredibly unpopular in every poll (The Verge)

By Sean Smith · AI Bandwagon