America’s new AI map shows something surprising: ‘A lot of normal people are adopting AI’

Microsoft’s U.S. AI Diffusion Report, released Tuesday, suggests that picture is badly incomplete. Juan Lavista Ferres, Microsoft’s chief data scientist and the lab director behind the report, said within his own company, lawyers are building tools—people who are not software developers are translating their ideas into applications. Now that’s a big tech company where people are being actively encouraged to adopt AI tools, but he told Fortune he was surprised by his AI map: “A lot of normal people are adopting AI.”

The data backs him up—and the geography surprised even him.

Texas comes out ahead of California

The report—which tracks AI user share across all 50 states, the District of Columbia, and more than 3,100 counties—puts Texas fourth nationally at 35.4%, ahead of California at 34.1% and New York at 32.9%. The top of the leaderboard belongs to the District of Columbia (40.6%), Maryland (36.5%), and Utah (35.9%). Leaders cluster in the mid-Atlantic corridor, the Mountain West, and the Sun Belt; laggards sit in Appalachia, the Northern Great Plains, and rural New England, where West Virginia brings up the rear at 20.8%.

Lavista Ferres said he was genuinely surprised by California’s position.

“A lot of people would associate that as [the leader], the majority of the models are created in California,” he told Fortune. “But the fact that you have states like Texas or Utah or Maryland ahead of California was interesting for us.”

That Texas outranks California tracks with a broader demographic and economic realignment the Census Bureau has been documenting for years. The five fastest-growing cities in the United States are all in Texas, in the Dallas and Houston suburbs, to be exact. Also, the Houston and Dallas-Fort Worth metros added more residents last year than any other metros in the country.

On the ground, that migration is producing exactly the kind of AI-forward entrepreneurship the diffusion data captures: Fortune has reported on Fathom AI, an Austin-based sales platform built by a three-person team that launched in early 2026 with $300 in capital and reached $300,000 in annualized revenue within 12 weeks, driven almost entirely by AI agents handling tasks that would have previously required a full sales force.

Asked directly whether it was fair to connect Texas’s AI adoption and population growth, Lavista Ferres didn’t hesitate.

“I think it’s completely fair,” he said. “I think there is a connection. It’s difficult sometimes to talk about causality, right now we only need to talk about correlation, things that aren’t necessarily causal at this point, but I do think that there is a phenomenon there.”

Another urban-rural divide

The state-level numbers are striking. The county-level numbers are alarming.

Across more than 3,100 counties, Microsoft found AI use averages 33% in metropolitan areas, 22% in micropolitan ones, and just 16.2% in rural counties—a 16.8 percentage point gap between the most and least connected parts of the country. Critically, that gap persists after controlling for age, income, and demographic composition.

Lavista Ferres said this was “quite striking,” and even though people may cite older demographics or wealth inequality when discussing rural populations, “even controlling for all those factors, you still have a big gap.”

Citing his own upbringing in rural Uruguay, the Microsoft data scientist said the divide in the U.S. “is kind of the norm in in multiple countries where you see the the technological divide between the rural and urban areas.” He grew up in a country where that divide was simply assumed to be structural and permanent. The Microsoft data suggests the same dynamic is present in the U.S. as well.

The implications compound quickly. If AI adoption is a leading indicator of productivity and wage growth, the urban-rural divide isn’t just persisting—it may be accelerating, leaving the communities least connected to the AI economy also the least equipped to catch up. Lavista Ferres said Microsoft is already tracking the productivity connection at the micro level. He cited randomized control experiments—including a widely circulated Harvard/BCG study on consulting productivity—and offered a personal benchmark: A report that would have taken his team months to build was completed in a week using AI coding tools.

“The best software developer in the world cannot compete with the average software developer using these tools,” he said. “There’s no competition.”

That productivity dividend is already reaching beyond the tech industry. Fortune has reported on Rick Chorney, a 29-year-old high-school dropout running a janitorial services company in the suburbs of Vancouver, who used AI tools to triple his revenue to nearly $1 million in a single year by automating customer intake, installing an AI receptionist handling 15 calls an hour, and compressing what once took years of costly trial and error into a matter of months. Chorney’s story is a ground-level example of what the Microsoft diffusion data captures in aggregate: AI adoption spreading through small and medium-sized businesses far outside the traditional technology corridor.

College towns are the hottest AI markets

The top AI-using county in the country isn’t in Silicon Valley. It’s Williamsburg, Va.—home to the College of William & Mary—where AI user share hits 73.7%. Harrisonburg, Va. (James Madison University) follows at 67.9%, then Madison, Idaho (BYU-Idaho) at 67.7%, Brazos County, Texas (Texas A&M) at 64.5%, and Story County, Iowa (Iowa State) at 64.2%.

This was a surprise to Lavista Ferres—and in fact, he didn’t even see it at first.

“We were doing an analysis on the top 20 counties—just looking at the list—and [someone on my team] said, ‘These are college towns.’ And that’s when we start going like, ‘Okay. There’s something happening in the college town that is different than the rest.'”

He noted that if Williamsburg were a country, its AI user share would rank first in the world. Counties where more than 10% of the population is aged 18–24 average 28.8% AI user share, versus 20.5% everywhere else.

Lavista Ferres said he has data on what happens to college towns during the summer, and the younger age-bracket population usually comes down when students leave.

“We want to continue doing a deep dive on college towns,” he said.

New York is trailing the pack

At 32.9%, New York ranks 14th nationally, below not just California but Georgia, Massachusetts, Connecticut, Illinois, and Rhode Island. For a state that houses the country’s largest financial sector and a significant share of its technology industry, the gap between reputation and data is at least worth examining.

Lavista Ferres was careful not to overread it.

“I’m not saying it’s not there,” he said, and there are some big cities that perform well, but his team hasn’t done a deep analysis on that yet. State-level AI user share can mask significant intra-state variation; a high-adoption metro like New York City could theoretically be pulling the state figure up even as surrounding areas drag it down, or vice versa. He said he hopes future reports will include deeper metro-level breakdowns.

With regard to the housing market, to return to the Texas comparison, New York is trailing. New York City lost 12,196 residents last year, the largest numeric population decline of any city in the country. The Northeast’s largest cities went from 1.2% average population growth to 0.2% in a single year.

The geographic pattern may partly reflect something beyond infrastructure and industry mix: attitude. This year’s Axios Harris Poll 100, also released Tuesday, finds 44% of Republicans say their opinion of AI has grown more positive in the past year, compared with just 35% of Democrats. The states outperforming in AI adoption—Texas, Utah, Nevada, Georgia—are among the most Republican in the country.

The partisan gap is sharpest around specific companies. OpenAI’s reputational score was just one point higher among Republicans than Democrats in 2024; today that gap has widened to 12 points.

“The cultural fault lines are quickly being drawn on whether AI is a benefactor or a ‘broligarchy,'” John Gerzema, CEO of The Harris Poll, said in a statement accompanying the data. The Microsoft diffusion report measures behavior, not sentiment—but the two datasets, read together, suggest adoption and attitude appear to be moving in the same direction, along the same political geography.

What the map means—and what it doesn’t

Lavista Ferres was measured about what the data can and cannot yet prove. But he was optimistic about what the diffusion of AI beyond elite corridors signals. Inside Microsoft, he said, a lawyer with dyslexia recently showed him a tool he was building with AI.

“He was basically building tools to help him and not only was the tool great, he showed it to the Windows team and they said, ‘We’ve actually been thinking about something like this for a long time. We might get some of your ideas.'”

He said he was optimistic about some kind of “renaissance” because of examples like this: “What will matter the most is these tools will help you get an idea and make it to production in a much easier way.”

The entrepreneurs Fortune has covered in recent months—from Chorney’s janitorial business in suburban Vancouver to Fathom AI’s three-person medical aesthetics platform in Austin—reflect the pattern Lavista Ferres describes. The technology is not staying in the lab. It is moving into the subdivisions, the college towns, and the small businesses of places that, until recently, were watching the AI economy from the outside.

The American economy is not adopting artificial intelligence uniformly. It’s adopting it along the same fault lines—density, education, employer mix, infrastructure, and increasingly, politics—that have structured economic inequality for decades. The difference now is that AI may be widening those fault lines faster than any previous technology wave. Microsoft plans to release updated diffusion data every three months.

For this story, Fortune journalists used generative AI as a research tool. An editor verified the accuracy of the information before publishing.