Seven in 10 Americans oppose a data center in their area

Monday morning, on stage at the All-In Summit in Los Angeles, Jensen Huang took a call he couldn't send to voicemail. It was Donald Trump. Nvidia's CEO put him on speaker for the room, and the president disposed of the moment's biggest debate about the pace of artificial intelligence in a single line: "we're not going to let that happen. It's a hoax."

TechCrunch told the story that evening. The reporter goes on to note that Trump and his allies cast hostility toward data centers as a foreign influence operation, and answers it with a number: "Recent Gallup polling shows that seven in 10 Americans oppose the construction of data centers in their area."

The number checks out. The word in front of it doesn't.

What the poll asked

Gallup's question sits in plain sight in the public document, in its exact wording: "Overall, would you strongly favor, somewhat favor, somewhat oppose, or strongly oppose the construction of a data center in your area to support artificial intelligence, or AI, technology in the U.S.?"

The answers fit on a single row of a table. Seven percent strongly favor, 20% somewhat favor, 23% somewhat oppose, 48% strongly oppose, 3% no opinion. That adds up to 27% in favor, 71% against, with a margin of sampling error of ±4 points.

The same field period carried a neighboring question about a nuclear power plant. Result: 53% opposed. Gallup has tracked that item since 2001, and its record high for opposition is 63%. So in March, a data center drew more refusals than nuclear power ever has.

The word that doesn't fit

The methodology note runs below the article, and it leaves nothing to guesswork: "telephone interviews conducted by Recon MR March 2-18, 2026, with a random sample of 1,000 adults".

March 2-18 was six months ago. A poll is a photograph, not a window. It tells you what a thousand people thought at that moment, and nothing else.

That leaves the most generous reading: maybe Gallup released an older survey late, and "recent" points to the publication rather than the fieldwork. That reading doesn't hold either.

The page carries its publication date in four places, the PDF of the topline results is named after it, and the web archive holds twelve captures from May 13 and 14. The poll came out on May 13, 2026: the release is four months old, the fieldwork six.

The date survived a copy-paste

On the day it landed, the Washington Post ran a story whose second sentence dated the whole thing: "The poll, conducted this spring and released Wednesday". Field dates and release, in eight words.

Four months later, on September 12, The Prowers Journal, a free local news site in Lamar, Colorado, republished Gallup's release word for word, author byline included. The sentence "These results, from a March 2-18 Gallup survey" is still sitting there, untouched.

So the date made it through a wholesale copy-paste at a county outlet. It disappeared forty-eight hours later, in a sentence written to summarize.

One sentence, two surveys

The rest of that same sentence adds the reasons: "more than 50% of respondents citing the effect of data centers on environmental resources", then "about 20%" for cost of living and quality of life.

Those numbers aren't from the March survey. Gallup says as much in its own text: the motivations come from "an open-ended question on an April web survey using the Gallup Panel". The methodology note fills in the rest: self-administered web surveys, April 1-15, 2,054 adults drawn from the panel, with subgroups of 472 supporters and 1,561 opponents.

Different field dates, different mode, different sample. Two photos taken a month apart on two different cameras, slipped into the same frame. The frame makes them look like one shot.

At the bottom of its page, Gallup offers a link labeled "View complete question responses and trends". We downloaded it: three pages, the March telephone questionnaire, its crosstabs, its methodology.

The April survey appears nowhere in it. Not the word April, not the word Panel, not the 472 and 1,561 subgroup sizes. That's consistent, since the document is the record of the phone survey, but a reader who clicks through to check the reasons won't find them there.

Opponents turned into respondents

The third gap is the quietest of the three, and it's a change of population. Gallup writes "Half of opponents mention data centers' excessive use of resources". The pickup writes "more than 50% of respondents". Half of a group is not half of the room: opponents account for roughly three quarters of the April sample.

The same slide happens on the rest of it. Gallup describes two separate groups of about a fifth each, one worried about quality of life, the other about economic consequences. The pickup merges them into one "about 20%".

One more detail rounds out the picture. In the TechCrunch piece, the hyperlink sitting on "environmental resources" doesn't point to Gallup, but to another TechCrunch story, published September 9, about new rules in Massachusetts. A reader who clicks to check the number lands somewhere else entirely.

What Gallup got right

None of the above would have been checkable if the firm had worked any other way. Gallup dates its fieldwork to the day, names the vendor that placed the calls, publishes its margin of sampling error, gives the exact wording of its question, puts its topline online, and flags itself that the motivations come from a second survey. It left the labels on the boxes.

And a March poll is still a valid poll. For March.

The measurement that was missing

There is a more recent measurement, and it's easy to reach. On August 19, Heatmap published a survey run by Embold Research from August 8 to 13 among 2,045 registered voters, margin of ±2.3 points: 75% opposed to a data center near where they live, with more than six in ten Americans strongly opposed.

Their series is instructive because the question wording hasn't moved across four waves. In August 2025, opinion was split, 43% in favor and 42% against. In February, 51% against.

In May, seven in ten. In August, three in four. Thirty-three points in twelve months.

That survey is not Gallup's and doesn't stand in for it: different firm, different mode of interviewing, and above all a different population, registered voters rather than adults. You don't mix the two series.

Corrected, Monday evening's sentence would read something like this: a March poll found 71% opposed, and a separate survey, in August, finds 75. It would lose its adverb and gain an argument.