NextFin News - The White House and Silicon Valley's chief executives are selling artificial intelligence as a political winner and an economic inevitability. The American public is not buying it. A cluster of surveys released this month shows President Donald Trump's approval rating on AI handling underwater by 34 points, with only 23% saying he has handled the issue well, while 69% of voters oppose AI data centers being built in their own communities - including a majority of Republicans. The gap between the booster narrative and voter sentiment is the clearest political risk hanging over the AI trade as midterm elections approach.
The Numbers: A Public That Rewards Use, Not Rhetoric
The polling picture is unusually consistent across three independent surveys, and it points to a single conclusion: voters separate the technology they use from the politics of its buildout, and they judge the latter harshly.
The University of Massachusetts Amherst national poll, fielded August 21-26 among 1,000 respondents with a margin of error of plus or minus 3.5 percentage points, found that 23% of Americans believe Trump has handled artificial intelligence "very well" or "somewhat well" - 5% and 18%, respectively - while 57% say he has handled it poorly. The intensity is worsening: the share saying he has handled AI "not well at all" rose from 36% in March to 43% in August, with 20% undecided. CNN data analyst Harry Enten calculated the president's net approval on AI at 34 points underwater overall and 61 points underwater among independents.
The Rainey Center's September Policy Survey, conducted September 10-14 among 1,004 registered voters (margin of error plus or minus 4.9 points), adds a sharper cut. Daily AI users are the only constituency where Trump posts a net-positive approval rating: 54% approve of his performance to 45% who disapprove, a net of plus 9 points. Support collapses with usage frequency - weekly users approve 32% to 67% (net minus 35), monthly users 24% to 74% (net minus 50), and those who rarely or never use AI tools sit at 43/54 and 45/54 respectively. Overall, the survey puts Trump's approval at 41% with 59% disapproving.
That paradox - the heaviest users of the technology are the only group that rewards the president - matters because AI adoption is moving fast. The same survey found 52% of voters use AI tools at least monthly and 42% use chatbots. Yet 41% say nothing about AI excites them, and 70% are annoyed to learn content was AI-generated. Americans are adopting the tool while disliking what it represents.
The local dimension is even starker. The NBC News Decision Desk poll powered by SurveyMonkey, fielded August 20 through September 1, found 69% of respondents oppose the construction of AI data centers in their local area, with 45% "strongly oppose." Opposition is cross-partisan: 57% of Republicans, 81% of Democrats and 71% of independents are against data centers being built in their communities. The UMass Amherst poll put local support even lower - just 11% support construction in their community versus 65% opposed - and found that even among voters who describe themselves as "very much" MAGA, 42% oppose a local data center versus 25% who support it. Among 2024 Trump voters, 52% oppose local construction to 20% who support it.
And yet, for all the heat, AI remains a low-salience issue nationally. The Rainey Center found AI registers at just 12% as a top-three issue and data centers at 7%, while cost of living leads at 57% - more than double jobs and the economy at 30% - and 80% call inflation critical to their daily life. This is the shape of the political risk: not a national verdict, but a local veto.
The Mechanism: National Glory, Local Pain
The transmission channel is straightforward, and it is why the White House's framing keeps missing. The benefits of AI buildout accrue to capital owners and to the national accounts - GDP, productivity, corporate earnings. The costs land on specific communities: electricity, water, land use, grid strain.
Voters are responding to the bill, not the vision. Among Rainey Center respondents unfavorable toward data centers, the leading objection is electricity costs for residents at 28%, followed by water use at 18%, distrust of the companies at 14%, and AI job loss at 11%. Noise and appearance draw just 2%. This is not a NIMBY aesthetics fight; it is a utility-bill argument.
The water concern is widely held and media-transmitted: 48% believe a data center uses far more water than almost any other facility, and among them the sources are local TV and newspapers at 45%, national newspapers at 32%, and cable at 27% - with social platforms minor. Ray La Raja, co-director of the UMass Amherst poll, put the political consequence plainly:
Americans aren't giving President Trump particularly high marks for his handling of artificial intelligence. Just 23% say he has handled AI well, compared with 57% who say he has handled it poorly, and the public has grown more negative since March. The share saying Trump has handled AI 'not well at all' has risen from 36% to 43%. AI may be a new issue, but it is fast becoming one on which politicians are judged negatively.
Dawn Vogel, a Republican county supervisor in rural Buchanan County, Iowa - a county that favored Trump 63% to 35% in 2024 - voted earlier this year for a moratorium temporarily banning data centers. On the trade-off between data infrastructure and local resources, she said:
It's an issue of what's the best use for the land, what's the best use for our natural resources. You can have all the data you want to in the world, but if you don't have water to drink, you don't have humans to use the data.
The institutional response is already locking in. The Rainey Center found 71% support requiring community benefits agreements before construction, and 38% name making data center companies pay their own electricity a top AI priority for lawmakers. Forty percent say a candidate's opposition to data centers makes them more likely to vote for that candidate, against just 15% less likely. In a forced county-office choice, the opponent of a data center wins 44-20.
The China Argument Does Not Travel
Trump's public frame for AI is national competition. In a Truth Social post at the end of August, he wrote: "The only reason that communities throughout the U.S.A. should not want Data Centers is if they want to end up being backwards and poor. If they want to be successful and rich, with far lower taxes and jobs all over the place, let Data Reign." In a separate post he called opposition "a SICK conspiracy going on against AI and Data Centers, and the only one that is happy about it is China," adding, "WHOEVER WINS AI, WINS!"
Voters are not following him there. The Who Decides survey found only 18% think the United States falling behind China competitively is the main problem with AI, with most saying a lack of regulation is a bigger concern. The Rainey Center's May survey of 1,010 registered voters ranked "beating China in AI" dead last among seven data center policy priorities - only 10% ranked it first, 36% ranked it last. While 48% support a ban on storing US data in Chinese data centers, when forced to choose, 48% prioritize local concerns over building domestically, against 28% - a gap that has doubled since February.
The national-security frame works only at the margins. The local-cost frame wins. And the policy preferences are bipartisan: 78% support watermarking requirements for AI-generated content with no partisan gap - 82% of Democrats, 77% of Republicans, 75% of independents - and 61% want AI regulated mostly at the federal level for uniformity, against 20% preferring state-level variation.
The Market Is Pricing the Opposite Bet
While voters sour, the market has spent two years building a massive trade on AI power demand. The latest session prices show that trade already fracturing. Vistra trades at $140.67, down 2.01% on the day and roughly 36% below its 52-week high of $219.82, with a 50-day average of $147.46. Talen Energy sits at $292.30, down 0.34% and about 35% below its 52-week peak of $451.28, against a 50-day average of $333.61. Dominion Energy is at $63.57, down 1.12%, below its 50-day average of $68.01 and year high of $72.99.
The platform layer, by contrast, remains near the top of its range. NVIDIA trades at $222.27, up 1.34%, with a market capitalization of $5.38 trillion. Microsoft is at $493.78, down 0.80%, market cap $3.67 trillion. Alphabet is at $349.54, up 0.64%, market cap $4.23 trillion.
That divergence - the infrastructure names down roughly a third from their peaks while the chip and cloud giants hold their ground - is the market's first hint that political and regulatory risk is being priced into the physical layer before it hits the platform layer. The ballot box is part of that repricing. Ballotpedia tracks at least 16 local ballot measures related to data centers in 2026 across seven states: California, Florida, Illinois, Michigan, Nevada, Ohio, and Wisconsin. As of August 26, voters had decided four of them, and in each case a majority favored greater limits on data center development - including an 88%-12% approval of a prohibition measure in Monterey Park, California, in June.
Cyclical or Structural? A Constraint, Not a Blip
The critical analytical question is whether this is a cyclical fluctuation that will mean-revert or a structural constraint that will not. The evidence points to a structural political constraint overlaid on a cyclical market correction in the power trade.
The structural case rests on four pillars. First, opposition is cross-partisan - Republicans, Democrats and independents all net against local data centers, which means it cannot be solved by winning an election. Second, opposition rises with age and rurality, with 54% of rural voters strongly opposed in the Rainey Center's read; these are precisely the communities where new facilities site. Third, the objections are material - power and water - not transient sentiment. Fourth, the institutional response is already hardening into law: community benefits agreements, ratepayer protection pledges, local moratoria, and ballot measures that have so far gone four for four against development.
The cyclical case has real force too. AI issue salience is low at 12% and can be crowded out by cost of living. Information treatments move voters toward support - the Rainey Center found every information treatment it tested shifted respondents toward support, and the share believing AI will destroy jobs falls from 60% to 53% after hearing that 60% of today's jobs did not exist in 1940. And 41% say neither party has a plan to address the future of AI, with the rest nearly evenly split between Democrats at 32% and Republicans at 27% - the issue is not yet owned by either side.
But the cyclical factors describe the national salience, while the structural factors describe the local permitting reality. Voters will tolerate AI if the local externalities are priced and governed. They will not tolerate a "let Data Reign" laissez-faire frame. That is a durable constraint on the buildout, not a polling blip.
The Second-Order Trade Nobody Is Pricing
The market is focused on the power trade - how many gigawatts, which utilities, which interconnection queues. The second-order move is the regulation trade, and it is barely priced at all.
The political equilibrium emerging from these surveys is not "stop AI." It is a specific, bipartisan package: federal rules with state-level uniformity, watermarking requirements, community benefits agreements, and making operators pay their own power. That package could pass. And if it does, it raises the cost base of the AI buildout while legitimizing it.
The second-order implication is counterintuitive: the companies best positioned are not the ones betting on deregulation. They are the ones that can absorb compliance costs, sign community benefits agreements, and lock in clean power at scale. The regulatory moat becomes a competitive moat. Capital-intensive incumbents with balance sheets thick enough to front the compliance and community-benefit costs gain an advantage over smaller, faster, less-capitalized developers who built their business case on cheap permitting and cheap power.
There is also a third-order expectation gap. The market has priced AI dominance as a national-security imperative that will override local objections. The voters are pricing it as a local externality that will be regulated, compensated, and slowed. If the voters are right, the terminal impact is not a halt to AI - it is a higher cost of capital, a slower buildout map, and a re-rating of which layer of the stack captures the returns.
The Strongest Counter-Thesis
The strongest argument against this reading is that AI is a low-salience issue and the economy is what drives midterms. The Cato Institute, in collaboration with Morning Consult, surveyed 4,150 registered voters August 13-16 with a modeled margin of error of plus or minus 2 points and found 75% say tariffs will be important to their vote this year - 35% "very important," 40% "somewhat important." If cost of living and tariffs dominate the fall, AI opposition stays a local nuisance rather than a national verdict. And the one group that rewards Trump - daily AI users - is the fastest-growing demographic in the electorate, with 52% using AI at least monthly.
This counter-thesis is right about the national level and wrong about the local level. AI will not decide the presidency. But it will decide county supervisor races, state legislature seats, and ballot measures - and those are exactly the venues where data center permitting happens. The constraint is local, and it is binding. A developer does not need the national electorate to approve its project; it needs a county board, a state utility commission, and a local grid operator. Every one of those is more exposed to a mobilized local majority than the White House is.
What Would Prove This Wrong
The falsifying signal is specific and observable. Watch the next Rainey Center or UMass Amherst read on data center favorability and priority trade-offs. If data center favorability rises above 40% - from its current 29% in the Rainey Center's September read - AND the share prioritizing local concerns over domestic buildout falls below 40% in two consecutive polls, the structural-constraint thesis is wrong and the "let Data Reign" frame is working. Also watch the remaining 12 local ballot measures this year: if any passes a pro-development outcome, the four-for-four pattern breaks and the local veto weakens.
Who Benefits, Who Is Exposed
The beneficiaries of the emerging regulatory equilibrium are clear. Utilities and generators that can sign long-term clean-power deals and pass through costs under ratepayer protections win. Large-cap cloud and chip names with balance sheets thick enough to absorb compliance and community-benefit costs win. Candidates who run against data centers in swing districts win - 40% of voters say opposition makes them more likely to support a candidate, against 15% who say less likely.
The exposed are equally clear. Independent power producers betting on a deregulated, fast-permitting environment face a deteriorating political backdrop. Local Republicans in data-center-heavy districts who toe the White House line face a primary and general-election crossfire. And the "beat China" rhetorical frame - which ranks last among voter priorities - loses persuasive power with every poll.
Outlook: Three Horizons, Three Scenarios
The forward view splits by time horizon, and the horizons point in different directions.
In the short term - the weeks to the November midterms - AI stays low-salience. Cost of living, inflation, and tariffs dominate the national conversation, and market volatility in the power names continues as the regulatory picture stays unclear. The issue is a background risk, not a headline driver.
In the medium term - six to eighteen months - ballot measures and state legislatures set the rules. Expect a patchwork: permissive states that court data centers with tax breaks and fast-track permitting versus restrictive states that impose moratoria, water limits, and community benefit mandates. That fragmentation fragments the buildout map itself, and developers will reroute capital accordingly.
In the long term - the structural horizon - the AI buildout proceeds, but under a regulated, community-compensated model. The cost of capital rises. The winners are the capitalized and the compliant. The technology does not stop; the economics change.
Three scenarios frame the path. The base case, roughly 60%: AI remains a local rather than national issue; regulation arrives as a bipartisan federal floor plus state-level variation; the power trade stabilizes but does not re-rate to its 2025 highs. The upside case, roughly 20%: cost-of-living relief and a strong jobs print crowd AI out of the conversation entirely; the "let Data Reign" frame wins by default; power names re-rate higher. The downside case, roughly 20%: a data-center incident - a grid failure, a water crisis - nationalizes the issue; a federal moratorium or heavy federal siting rules arrive; the infrastructure layer reprices materially lower.
The White House is winning the AI argument in boardrooms and losing it in backyards - and in American politics, the backyard usually votes first.
Explore more exclusive insights at nextfin.ai.
