Siena AI Poll Finds Broad Data Center Opposition, but Neither Party Owns the Issue

The Siena AI poll gives Republicans a 42 percent to 40 percent edge over Democrats on handling artificial intelligence. Yet that two-point difference does not establish political ownership of AI. It sits within the poll’s sampling uncertainty, while many voters still lack a clear view of either party’s position.

The stronger finding concerns physical infrastructure. Sixty-one percent of likely voters oppose constructing data centers to support AI, compared with 34 percent who support it. However, fewer than one percent identify AI or data centers as the most important issue affecting their November vote.

Those results create a striking political split. Voters express substantial concern about AI infrastructure, but that concern has not become a defining national voting issue. The immediate pressure falls on candidates defending local projects, not necessarily on either party’s national campaign.

What the Siena AI Poll Actually Found

The poll reveals intense opposition to AI infrastructure without a clear partisan winner on AI policy.

The New York Times and Siena College surveyed 1,503 likely voters nationwide from September 8 through September 13. The resulting New York Times report on the Siena AI voter poll examined party trust, data center construction, and the issues influencing midterm votes.

When asked which party they trusted more to handle AI, 42 percent chose Republicans and 40 percent chose Democrats. That narrow difference was within the poll’s margin of error, so it should not be treated as a meaningful Republican lead.

The remaining share is more revealing than the two-point gap. Eighteen percent did not select either party, a much larger undecided group than appears on established political issues.

New York Times political analyst Nate Cohn noted that 82 percent expressed a party preference on AI. By comparison, 97 percent chose a party when asked about immigration. The difference suggests that voters recognize AI as a concern without knowing which party represents their preferred response.

Age and gender produced sharper divides. Republicans led Democrats by 15 points among voters aged 45 to 64. Democrats led by 32 points among voters aged 18 to 29.

Men favored Republicans on the issue by 14 points, while women favored Democrats by six points. These differences show that an overall 42-to-40 result can conceal substantial demographic variation.

The finding also stands apart from the poll’s other issue tests. Democrats led on the economy, cost of living, immigration, foreign conflicts, and the federal deficit. AI was the only tested subject where Republicans held even a nominal advantage.

That exception does not mean voters endorse the Republican Party’s current AI agenda. Party trust questions measure a broad impression, not agreement with every policy or project. Respondents can prefer one party on AI while opposing infrastructure promoted by that party’s leaders.

The data center result was much less ambiguous. Sixty-one percent opposed construction, while 34 percent supported it. The question described the facilities as data centers built to support artificial intelligence technology in the United States.

Opposition crossed partisan lines, although its intensity varied. Seventy-five percent of Democrats opposed construction, along with 60 percent of independents and 47 percent of Republicans.

Republicans were almost evenly divided. Forty-nine percent supported construction, only two points above the share opposing it. That split complicates national messages centered on rapid infrastructure development.

Age produced an even more dramatic pattern. Nearly 80 percent of voters under 30 opposed data center construction. Opposition declined with age, but 53 percent of voters aged 65 or older still opposed it.

The Siena AI poll therefore records two distinct kinds of uncertainty. Voters remain uncertain about which party should manage AI, but they express much less uncertainty about unrestricted data center construction.

That distinction matters because “AI policy” covers several different concerns. It can mean job displacement, model safety, national security, competition with China, electricity demand, water consumption, or local land use.

A voter might support American AI research while opposing a nearby server campus. Another might distrust advanced models but accept a facility that funds its own energy supply. A single party-trust question cannot separate those positions.

The poll’s most important contribution is not a partisan horse race. It is evidence that voters distinguish between AI as an abstract national technology and data centers as visible local infrastructure.

Data Center Opposition Is Broader Than a Simple Ban

The 61 percent opposition figure reflects demand for limits and enforceable conditions more than unanimous support for prohibition.

Among respondents opposing data center construction, 56 percent preferred limits instead of an outright ban. That finding makes the overall result more nuanced than a headline about widespread rejection.

The distinction creates room for conditional approval. Voters can oppose the current pace or terms of construction without rejecting every future project. Conditions involving power costs, water, land, and public oversight can therefore change the debate.

Climate and environmental concerns were the most common explanation among opponents. Nearly one-third of that group cited those issues, although priorities differed by party.

Democratic opponents were almost twice as likely as Republican opponents to mention environmental worries. Republicans more often expressed general distrust of AI itself. Eighteen percent of all opponents cited broad AI distrust rather than a specific local impact.

Separate polling supports the conclusion that concern extends beyond one survey. An Associated Press-NORC Center for Public Affairs Research and University of Chicago Energy Policy Institute poll found that about six in ten Americans favored limiting new data centers. Most Democrats and Republicans supported some restrictions.

According to the Associated Press report on that environmental concern survey, 53 percent were extremely or very concerned about AI’s environmental effects. The comparable figure was 41 percent one year earlier.

Majorities also expressed high concern about effects on local electricity prices and water supplies. About two-thirds supported requiring data centers to obtain their electricity from clean energy sources.

Those concerns have a measurable physical basis, although local effects vary considerably. A small cloud facility and a hyperscale AI campus do not impose the same demands.

Hyperscale data centers are large computing campuses designed to operate thousands of servers at enormous scale. AI training and inference can add unusually dense and sometimes variable electricity loads.

A 2026 Lawrence Berkeley National Laboratory update estimated that data centers could consume 11.8 percent of total US electricity in 2030. Its reference case projected 649 terawatt-hours, with scenarios ranging from 521 to 843 terawatt-hours.

The Lawrence Berkeley National Laboratory electricity usage report places those estimates between 9.5 and 15.3 percent of national consumption. The range reflects uncertainty about equipment shipments, chip use, server utilization, and operating life.

National consumption figures do not determine what happens to a particular utility bill. Rate structures, grid capacity, generation contracts, and required upgrades differ across utilities and states.

The local scale can still be consequential. Large projects may require new generation, substations, transmission lines, or water systems. The central political question is who pays for those investments and who carries the risk if projected demand changes.

Developers and industry groups emphasize jobs, construction spending, tax revenue, and digital services. Dan Diorio of the Data Center Coalition told the Associated Press that the industry provides tangible benefits in communities nationwide.

Those benefits are not automatically distributed evenly. Construction can create substantial temporary employment, while ongoing operations may require fewer workers. Tax benefits depend on local agreements, incentives, and the public services required by a project.

The geography of expansion adds another source of tension. Pew Research Center identified more than 3,000 operating US data centers and over 1,500 projects in development as of February 2026.

Its data center analysis found that 87 percent of operating facilities were in urban areas. In contrast, 67 percent of planned facilities were located in rural areas.

Thirty-nine percent of planned projects were in counties without an existing data center. That means many communities will confront unfamiliar infrastructure decisions without decades of local experience.

Rural sites can offer available land and easier access to energy projects. They may also have smaller water systems, limited transmission capacity, and local governments with fewer technical resources.

The shift explains why data centers can become politically important without ranking highly in national issue surveys. A facility might remain abstract to most voters while becoming central within the county considering its permits.

This is the mechanism behind broad AI data center opposition. National anxiety becomes politically actionable when residents see a proposed site, utility plan, tax agreement, or water allocation.

Neither Party Has Turned Concern Into a Coherent Position

The central contest is not Republicans against Democrats, but national AI ambition against demands for local control.

President Donald Trump has promoted AI development and the infrastructure needed to support it. The administration’s official AI Action Plan connects faster data center development with investment, national security, and competition with China.

That position gives voters a clear statement about national ambition. It offers less clarity about how individual communities should evaluate power, water, noise, land, and utility costs.

The Republican coalition does not hold a single view. Some candidates favor rapid development, while others call for restrictions or stronger oversight. Republican voters themselves are divided almost evenly on construction.

Democrats also lack a unified national position. Progressive lawmakers have demanded stronger AI controls, while state and local Democratic officials have sometimes supported projects promising investment and tax revenue.

The disagreement crosses party lines because the national and local incentives point in different directions. Federal leaders want computing capacity and technological leadership. Local officials must answer questions about permits, utility bills, and community resources.

Candidates have started adjusting their language accordingly. Republican pollster Tony Fabrizio warned that unconditional support for more data centers was politically damaging.

His reported candidate guidance urged Republicans to validate concerns before supporting projects. It recommended requiring developers to protect ratepayers, supply their own power, and safeguard water resources.

Fabrizio’s separate polling found that 61 percent supported construction with conditions, while 28 percent favored prohibition. Respondents opposed additional construction without conditions by 65 percent to 24 percent.

Those numbers came from a different survey and should not be merged directly with the Times/Siena results. They nevertheless reinforce the importance of how questions describe project conditions.

A candidate saying “build more data centers” asks voters to trust developers and regulators before project details are known. A candidate demanding enforceable conditions offers a different proposition.

This creates a political challenge for both parties. Neither can simply label itself pro-AI or anti-AI and expect that position to match the public’s fragmented concerns.

The AI midterm issue also lacks stable partisan cues. On older subjects, voters often know what Republican and Democratic labels imply before hearing specific proposals.

AI combines priorities that do not fit neatly into those established divisions. National security arguments favor faster investment. Labor concerns can favor restrictions or worker protections.

Environmental concerns encourage scrutiny of energy and water use. Economic development interests encourage construction. Technology safety concerns can support model regulation without determining where infrastructure should be built.

The Times/Siena numbers capture that unsettled alignment. Republicans receive 42 percent trust even though the administration strongly favors expansion. Democrats receive 40 percent despite greater opposition to data centers among Democratic voters.

Neither side has converted its position into broad political ownership. The 18 percent without a party preference remains evidence of that gap.

For technology companies, this uncertainty is not neutral. It raises the cost of winning local approval and increases pressure to disclose project impacts earlier.

Statements about jobs or national competitiveness are unlikely to answer every concern. Communities increasingly want information about total power demand, backup generation, expected water consumption, noise, tax incentives, and long-term employment.

The Department of Energy has acknowledged the need to protect households and businesses from undue costs while expanding AI infrastructure. Its official data center resource hub says technology companies participating in the administration’s ratepayer-protection framework should provide or procure new power, pay for required delivery upgrades, and negotiate separate rate structures.

Those goals sound compatible, but implementing them requires specific contracts and regulatory decisions. A general federal promise cannot determine how one utility allocates upgrade costs.

This is where local approval standards become the real opponent of unrestricted national expansion. The debate moves from whether America should lead in AI to which projects satisfy enforceable community conditions.

Companies that treat public concern as a messaging problem risk missing this change. Voters are asking material questions about who pays, who benefits, and what resources a facility consumes.

Candidates face the same test. Reassurance without a policy mechanism leaves the conflict unresolved. A workable position must explain how projects will be evaluated and rejected when their costs exceed their local value.

Why Strong Opposition Is Not Yet a Defining Midterm Issue

Voters can hold a negative opinion intensely without using it as the main criterion for a national vote.

Only 11 respondents among more than 1,500 named AI or data centers as the most important issue determining their November vote. That placed the subject below major concerns such as the economy and immigration.

The finding does not invalidate the 61 percent opposition figure. It measures a different dimension of public opinion.

One question asks whether voters support a type of infrastructure. The other asks them to choose a single priority from every issue affecting the election. A voter can oppose data centers while still voting mainly on household costs, health care, immigration, or another concern.

The “most important issue” format also sets a demanding threshold. It does not capture secondary concerns, candidate-specific reactions, or local projects that affect only part of the electorate.

This explains why the AI midterm issue can matter in selected races without dominating the national campaign. Its influence is likely to be geographically concentrated around proposed facilities, utility disputes, and visible construction.

National toplines can obscure that concentration. A data center proposal might affect one county deeply while remaining irrelevant to voters elsewhere in the same state.

The poll’s timing creates another limitation. Interviews ended September 13, and attitudes around AI policy were changing rapidly. Polls record views during their field period, not permanent public commitments.

The question wording also matters. Respondents were asked about constructing data centers to support AI technology in the United States. Their answers may combine attitudes toward AI, industrial construction, resource use, and nearby development.

The survey did not ask about AI safety. It therefore cannot establish whether model-related warnings caused the observed opposition or how voters would prioritize safety regulation.

It also cannot show how support changes when a project includes protections. Requirements involving dedicated generation, water recycling, ratepayer safeguards, or community payments can produce different responses.

The broad category of “data center” presents another challenge. Facilities vary greatly in size and design. Some support AI, while others host conventional cloud computing, financial services, streaming, medical systems, and business applications.

Many campuses support multiple workloads. Public debate often uses AI data centers as a general label even when the precise computing mix remains unclear.

Opposition can also reflect dissatisfaction with the approval process. Residents may object to limited disclosure, tax incentives, zoning changes, or insufficient public participation rather than the technology alone.

These qualifications do not erase the result. They define what the result can support.

The Siena AI poll supports the conclusion that unrestricted construction faces broad resistance. It does not prove that 61 percent favor stopping every proposed facility.

It shows no statistically clear party advantage on AI. It does not prove that both parties have equally trusted policies within every demographic group.

It finds extremely low first-priority salience nationally. It does not prove that AI infrastructure lacks electoral consequences in communities hosting major projects.

That final distinction is especially important. Political salience often emerges through repeated local experiences before it consolidates into a national partisan issue.

Electricity bills, water restrictions, zoning hearings, and construction disputes give an abstract subject practical meaning. A sequence of local conflicts can eventually change how voters interpret national party positions.

The opposite outcome is also possible. Better project standards and clearer cost allocation could reduce opposition before AI becomes a stable partisan dividing line.

Developers have incentives to make that case through verifiable commitments. Closed-loop cooling, dedicated generation, flexible electricity demand, and guaranteed rate protections can address specific concerns when technically feasible.

Communities still need independent ways to evaluate those claims. Voluntary company promises do not carry the same force as utility tariffs, permit conditions, public reporting, or enforceable contracts.

The skeptical reading is therefore straightforward. The poll captures real dissatisfaction, but it does not reveal one policy solution or a durable national coalition.

Anyone treating the 61 percent result as a mandate for prohibition goes beyond the evidence. Anyone dismissing it because AI ranks below one percent as a top issue makes the opposite error.

Both findings can be true simultaneously. AI infrastructure is broadly unpopular and nationally secondary, at least during the poll’s September field period.

Three Signals Will Show Whether the Backlash Lasts

The next phase depends on local elections, enforceable infrastructure rules, and whether either party builds a recognizable AI position.

The first signal is how candidates discuss active data center projects during the final weeks before the midterms. Generic support for AI will matter less than responses to specific community disputes.

Watch whether candidates demand that developers finance power generation and grid upgrades. Also watch for positions on water disclosure, zoning authority, noise standards, and tax incentives.

If those conditions spread across both parties, the poll’s main lesson will be strengthened. It would show that local opposition has changed the approval standard even without becoming a top national voting issue.

If candidates continue supporting projects without conditions and face little electoral pressure, the 61 percent figure will look less politically actionable. That outcome would highlight the difference between survey opinion and voting behavior.

The second signal is what utilities and state regulators require from large new loads. Data centers become political when households fear that infrastructure costs will appear on ordinary electricity bills.

Special rate classes, minimum purchase commitments, upfront infrastructure payments, and long contract terms can shift more risk toward developers. Public proceedings can also clarify the assumptions behind projected demand.

The Department of Energy’s transmission work describes data centers as one contributor to accelerating electricity demand. It also stresses coordination among communities, states, utilities, and grid organizations.

Policy details will determine whether that coordination protects existing customers. A pledge to prevent cost shifting matters only when regulators translate it into enforceable rate and interconnection rules.

If regulators broadly adopt such protections, conditional public support may grow. If household bills rise alongside data center construction, opposition will receive a concrete economic foundation.

The third signal is whether one party develops a stable, understandable national AI platform. The Siena AI poll shows that neither has done so yet.

A coherent platform would connect model safety, jobs, national security, energy infrastructure, and local consent. It would also explain which decisions belong to federal, state, and local authorities.

Republicans currently hold only a nominal 42-to-40 advantage. Democrats oppose data center construction more strongly, but that position has not produced a clear trust advantage on AI.

Either party could narrow the 18 percent uncertainty gap with specific policies. Neither can assume that voters will infer those policies from general statements about innovation or regulation.

Future polling should therefore be judged on more than the party-trust topline. The most useful indicators will include undecided responses, support under specified project conditions, and changes within communities hosting proposed facilities.

Readers should also separate national opinion from local exposure. Pew found that 42 percent of Americans lived within five miles of an operating or planned data center. Yet proximity alone did not substantially change public views in its survey.

That suggests awareness, approval procedures, and perceived cost allocation may matter more than distance by itself. A nearby facility can remain politically quiet when benefits and protections appear credible.

For developers and enterprise AI buyers, the lesson extends beyond electoral politics. Compute capacity now depends on public infrastructure decisions that affect timelines, costs, and location choices.

In practice, a software team may never attend a zoning hearing, yet a delayed grid connection can still mean fewer cloud accelerators in its preferred region, longer capacity reservations, or higher computing costs. Those constraints can reach companies that never participate directly in a local permitting dispute.

Knowledge workers and everyday AI users should also understand the connection. Every model request relies on physical facilities, electricity systems, cooling equipment, networks, and local approvals.

The Siena AI poll makes that infrastructure visible. It shows that voters are not automatically accepting expansion simply because AI services are widely used.

At the same time, the poll offers no evidence that AI will decide the national midterms. Fewer than one percent naming it as their top issue is a strong warning against inflated political conclusions.

The sharper interpretation is more useful. AI has become a substantial infrastructure concern before becoming a settled partisan issue.

That gap gives communities, companies, regulators, and candidates room to shape what comes next. It also means today’s 42-to-40 party split should be treated as provisional.

Watch the conditions attached to permits, utility agreements, and campaign commitments. Those decisions will reveal whether AI data center opposition becomes durable policy or remains broad but secondary dissatisfaction.

The central question is no longer whether Americans recognize AI’s physical costs. The poll suggests many already do. The question is whether institutions can address those costs before local opposition hardens into a national political identity.