Start by conceding the number, because it is real and it is not close. The June UT/Texas Politics Project Poll found 29% of Texas voters support building a data center in their community and 56% oppose it. The opposition is not a rural pocket. It is 62% in rural areas, 60% in the suburbs, and 45% in cities. It crosses party lines: 71% of Democrats, 62% of independents, and a Republican electorate split almost evenly at 44% opposed to 42% in favor.
Anyone who builds in Texas should take that seriously. I do. And it is no longer only an attitude. In June the governor directed regulators to make data centers fully fund the electric infrastructure built to serve them. On August 3 he went further, ordering the Public Utility Commission and ERCOT to audit every data center in the interconnection queue and pausing new approvals until they do (The Texas Tribune). The poll is now policy.
So the question of why those numbers look the way they do is not academic. It determines what gets built, by whom, and on what terms. And the answer being circulated is, I think, the wrong one.
The poll never asked why
The Texas Politics Project's write-up argues that Texans' mixed feelings about artificial intelligence are a likely contributor to their opposition to data centers. The supporting evidence is that the two attitudes travel together: 49% of Texans expect AI to hurt the economy against 29% who expect it to help, and the groups most sour on AI are broadly the same groups most opposed to data centers.
To its credit, the write-up says plainly what it is doing: "the June poll did not include direct 'why' questions for data center opposition." The causal story is an inference drawn from two measurements pointing the same direction. That is a reasonable hypothesis. It is not a finding. And the hedge tends to fall away as the claim gets repeated.
Two things rising at the same time is not one of them causing the other.
A simpler reading, from the same poll
There is a cleaner test available inside the survey, and the survey supplies both halves of it.
If attitudes toward AI were driving opposition to data centers, opposition should be distributed the way exposure to AI is distributed. It is not. Awareness of AI is close to uniform across the state — the write-up notes news about AI is reaching Texans "relatively uniformly," with roughly 60% in every kind of community saying they have heard a lot about it. Awareness of data centers is emphatically not uniform: 49% of rural Texans have heard a lot, against 39% in the suburbs and 36% in cities.
Now line the opposition up against each. Against AI exposure, which barely varies, opposition varies a great deal. Against data center exposure, it tracks almost step for step.
Across rural, suburban, and urban Texas, awareness of AI barely changes, yet opposition swings 17 points. It tracks how much people have heard about data centers instead.
UT / Texas Politics Project Poll, June 2026. AI awareness ≈60% reflects the source's description of “relatively uniform” statewide exposure.
A variable that is flat cannot explain an outcome that is not. Whatever is moving these numbers moves with how much people have heard about the facility, not about the technology inside it. That is exactly the pattern you would expect if the objection is to a specific project on specific terms rather than to a category of software.
The demographic detail points the same way. The sharpest divide in the poll is not between AI optimists and pessimists. It is inside a single party: 52% of Republican men support data centers in their community while 50% of Republican women oppose them, with only 31% in support. Among suburban women, 65% are opposed and 18% in favor. Same party, same news environment, broadly similar views on AI in the abstract. What differs is proximity to the household ledger.
I want to be careful here, because I am doing a version of the thing I just criticized. This is also an inference. The difference is that it is an inference about variables the poll actually measured, and it makes a prediction the AI-sentiment story does not: ask why, and you will hear about rates, water, traffic, and abatements far more than you will hear about chatbots.
The write-up itself lists the direct reasons people give — anticipated higher energy costs, environmental concerns, quality of life, and objections to state subsidies. Every item on that list is concrete, local, and financial. Not one requires an opinion about artificial intelligence.
One word for two different things
The deeper problem is that "data center" is a single phrase on a survey line covering two projects that have almost nothing in common.
The first connects to ERCOT and adds load to a system everyone else depends on. It joins a queue that now holds more than 1,800 projects representing over 474 gigawatts — more than five times the grid's record peak demand — roughly 90% of it data centers (Utility Dive). It takes a ten-, fifteen-, or twenty-year exemption from the state sales tax and commits, in the largest tier, to forty jobs.
About 1,800 projects are asking ERCOT for 474 gigawatts — roughly 90% of it data centers — against a record peak demand of 91 gigawatts. Not every project is real, but the gap is the point.
ERCOT large-load queue, July 2026; The Texas Tribune / Utility Dive.
The second brings its own generation, never enters the residential rate base, and pays full freight. Chevron and Microsoft signed a twenty-year deal for roughly seven billion dollars of turbines to produce 2.67 gigawatts for an AI data center near Pecos, in the Permian — entirely off-grid (CNBC).
A poll cannot distinguish those. Neither, fairly, can a respondent. Policy has to — and it just did. The governor's audit, by design, "doesn't apply to so-called 'behind-the-meter' projects that don't intend to connect to ERCOT." That single clause is now the most consequential sentence in Texas infrastructure, and it should be.
What changed on August 3
Abbott's directive freezes approvals and requires every queued project to disclose power and water use and sourcing, ownership and financing, reliance on tax breaks, and steps taken to mitigate noise, light, and traffic. Fail the audit and ERCOT denies the connection.
Every one of those disclosures is a question a well-structured project should already be able to answer on demand. The audit is not an obstacle to that kind of project. It is a filter that separates it from the rest of the queue.
Read it as a spec sheet
Once you stop treating the 56% as a mood and start treating it as a set of requirements, it resolves into four tests. Not one of them is unreasonable. Not one is about AI.
Power
Bring your own.A queue of 474 gigawatts against a grid whose record peak is a fraction of that is not a pipeline; it is a claim on capacity that does not exist. Senate Bill 6 already requires loads of 75 megawatts or more to accept curtailment during a grid emergency and to install the equipment that lets ERCOT act on it (Utility Dive). If a project only pencils when it can draw from the system everyone else depends on, it is not a neighbor. It is a competing claimant.
Price
Do not land on the neighbor's bill.This is the least understood item and the most damaging. ERCOT allocates transmission cost by Four Coincident Peaks — a customer's demand during a single fifteen-minute interval in each of four summer months. A large load that curtails through those four intervals can shed an enormous share of its transmission obligation. NRG puts the figure at $33.4 million a year for a 500-megawatt load, while the residential class carries roughly 48% of the allocation and pays it volumetrically, per kilowatt-hour (NRG, ERCOT Market Summit).
Transmission and distribution has already grown from 28% of an average residential bill at the start of retail competition to about 40% today. That is the bill people are voting against. It is a cost-allocation mechanism, not a sentiment about software.
Subsidy
No abatement dependence.A qualifying large data center in Texas needs 250,000 square feet, a $500 million investment, and forty jobs to earn up to twenty years of exemption from the 6.25% state sales tax (Texas Comptroller). Reasonable people can argue about whether that trade is worth making. What is harder to defend is how little it is checked: of 138 data centers in the program, only 20 have been audited, and 6 of those were breaching the terms — four never created the jobs they promised (The Texas Tribune).
Voters have already worked this out. Brazoria County commissioners voted down a tax abatement for a 620-megawatt project in March (The Texas Tribune). A project that needs the abatement to clear its own hurdle rate has told the county something important about itself.
Locality
Jobs, tax base, and a name on the door.San Marcos denied rezoning for a $1.5 billion facility over water. An East Texas project was withdrawn when the developer concluded it could not meet the state's new standards (KSAT). Abbott's audit now demands ownership and financing disclosure by name.
Communities are no longer asking only what is being built. They are asking who is building it, who is accountable when it goes wrong, and what stays behind when it is done. Those are the questions an owner can answer and a special-purpose vehicle usually cannot.
The tests are not a rebuttal
Nothing above is an argument against the 56%. It is a transcription of it. Read the four tests back to back and what they describe is not a hostile public. It is a public that has priced the deal accurately and declined the version currently on the table.
That distinction matters for anyone deciding what to build next. If the opposition is a diffuse cultural anxiety about artificial intelligence, there is nothing to do but wait it out or spend money on persuasion. If the opposition is a set of terms, the terms can be met — and meeting them is a design problem, which is a category of problem the industry is actually good at.
So why does the fight keep happening?
If the terms are knowable and the tests are meetable, the honest question is why this keeps ending in a no. I think it comes down to two failures, and neither one sits on the public's side of the table.
Education
Almost no one has been told how any of this works. Most people have never heard of a coincident peak, a behind-the-meter project, or an interconnection queue. They see the bill rise and a windowless building go up. When the mechanism is invisible, the only signal left is the outcome — and the outcome looks like a cost with someone else's name on the upside.
Candor
The industry has trained communities to expect evasion. Projects arrive behind a shell company with a code name. The power source is "still being finalized." The job count is the legal minimum dressed as a windfall. The water study comes after the vote. Each is a choice to disclose less than is known, and each reads — correctly — as something to hide.
You do not fix distrust with a better slogan. You fix it by removing the reasons for it. Education is answerable: explain the mechanism in plain words. Candor is answerable too: put the terms in writing before the vote, and say out loud what you will not do. What follows is an attempt to do exactly that. It is not a pitch. It is a format.
