AI Is Spreading Fastest Where Workers Already Earn the Most

Here is a number that should reshape how you think about AI and jobs: at firms with 250 or more employees, AI adoption tripled from roughly 4% in 2023 to about 12% by 2025. That is fast. But the part almost nobody is talking about is where that growth is concentrated — and it is not where the headlines told you to look.

A new analysis from the Peterson Institute for International Economics (PIIE), published on May 21, 2026 by Gary Clyde Hufbauer and Ye Zhang, digs into US Census Bureau survey data and finds something counterintuitive. The industries racing ahead with AI are the ones that already pay the highest wages and hire the most skilled workers. The industries everyone assumed would get automated first? They are the laggards.

If you have spent the last few years worried that AI would come for the lowest-paid, most routine jobs first, the data is telling a very different story.

The Sectors Pulling Ahead

By late 2025, more than 30% of firms in three specific industries reported they were already using AI: information, finance and insurance, and professional, scientific, and technical services. [Fact] These are not warehouses or call centers. They are the offices where analysts, engineers, consultants, and software developers do their work.

The concentration gets even sharper at the top. Among large information-sector firms with 250 or more employees, AI adoption is projected to reach roughly 73% by early 2026. [Estimate] In other words, in the corner of the economy that builds and sells information itself, AI is becoming close to universal among big players.

Now compare that to the laggards. Hufbauer and Zhang point to retail, accommodation, and food services as the slowest adopters. [Fact] These are exactly the industries that employ enormous numbers of lower-wage workers — the cashiers, the servers, the front-desk staff. The popular narrative for years has been that these jobs were first in line for automation. The actual adoption data flips that assumption on its head.

Why Bigger Firms Move First

Firm size turns out to be one of the cleanest predictors of who is using AI. The PIIE analysis shows that companies with 250 or more employees climbed from about 4% adoption to roughly 12% between 2023 and 2025. Smaller firms — those with 100 or fewer employees — moved from around 3% to about 8% over the same window. [Fact]

That gap matters more than it looks, because large firms are claiming a growing slice of all American jobs. Back in 2000, firms with 1,000 or more employees accounted for 37.6% of US employment. By 2025, that share had climbed to 42.3%. [Fact] So the businesses adopting AI fastest are also the ones employing a larger and larger portion of the workforce. The authors warn that faster AI adopters are likely to gain competitive advantages, which could push employment concentration in large firms even higher.

The Pattern: AI Is Following Skill, Not Replacing It

The thread tying all of this together is what economists call skill-biased technology adoption. Hufbauer and Zhang find a modest but real positive correlation between how much a sector's workers are paid and how quickly that sector adopts AI. [Fact] The higher the average compensation, the faster the AI uptake.

This is genuinely surprising, and it cuts against a lot of fear-driven commentary. The intuitive worry was that AI would be a great equalizer of destruction — that it would hollow out routine, lower-paid work first because those tasks are easiest to automate. Instead, in this snapshot of the US economy, AI is showing up first in the high-skill, high-wage service sectors. It is arriving as a tool in the hands of well-paid professionals before it arrives as a replacement for low-paid ones.

If you are a financial analyst, a software developer, a management consultant, or a scientific researcher, this means AI is already reshaping the daily texture of your work — not in some distant future, but right now. You can see how these high-exposure professions are tracked in detail on our financial analyst occupation page and software developer occupation page.

What the Data Does Not Say

It is worth being honest about the limits here, because PIIE itself is. The analysis draws on the US Census Bureau's Business Trends and Outlook Survey (BTOS), a Federal Reserve FEDS Note, an NBER survey of CEOs, and Bureau of Labor Statistics figures. [Fact] These tell us who is adopting AI. They do not yet tell us what AI is doing to employment.

On that question, the authors are deliberately cautious. They cite their colleague Jed Kolko, who concluded that the evidence so far on harmful labor impacts is inconclusive, and that claims about damage to particular groups of workers are premature. [Claim] Adoption is not the same as displacement. A finance firm using AI is not automatically a finance firm shedding jobs. The honest answer right now is that we are watching the input — fast, skill-concentrated adoption — without yet seeing a clear picture of the output.

The Global Picture Behind the US Numbers

It is tempting to treat this as a purely American story, but the dynamic Hufbauer and Zhang describe is showing up across advanced economies. Wherever AI tools are spreading, they tend to land first in the same high-skill service industries — information, finance, and professional services — because those are the sectors whose core work is information processing, the exact thing large language models are best at accelerating. A spreadsheet does not care which country it lives in, and neither does an AI assistant drafting a contract or summarizing a research report.

That global consistency is part of why the PIIE finding matters beyond any single labor market. If skill-biased adoption is a structural feature of how AI diffuses — rather than a quirk of one survey in one country — then the early concentration in high-wage work is likely to persist for a while. The cheapest, most routine, lowest-paid tasks are not necessarily the first to be touched, because the firms that employ those workers are often the slowest to adopt the technology in the first place.

What This Means for Your Career

If there is one practical takeaway, it is this: AI exposure is not the same as AI risk, and the two may be moving in opposite directions from what you would expect. The workers seeing AI arrive fastest are the higher-paid, higher-skilled ones in information, finance, and professional services. For them, the question is less "will I be replaced?" and more "how quickly can I learn to work alongside these tools before my competitors do?"

For workers in retail, hospitality, and food service, the slower adoption curve buys time — but it is not a guarantee of safety. Adoption laggards today can become adoption leaders tomorrow, especially as tools get cheaper and easier to deploy. The smart move in every sector is the same: treat AI fluency as a skill you build deliberately, the way an earlier generation learned spreadsheets and email.

The big-firm advantage is the part worth watching most closely. If the companies adopting AI fastest are also the ones employing a growing share of all workers, the gap between AI-fluent and AI-distant workplaces could widen quickly. Where you work may end up mattering as much as what you do.

Sources

• Gary Clyde Hufbauer and Ye Zhang, "The adoption of AI by industrial sectors," PIIE Realtime Economics, May 21, 2026. https://www.piie.com/blogs/realtime-economics/2026/adoption-ai-industrial-sectors
• Underlying data: US Census Bureau Business Trends and Outlook Survey (BTOS), Federal Reserve FEDS Note, NBER CEO survey, US Bureau of Labor Statistics.

---

AI-assisted analysis. This article was written with AI assistance and reviewed for accuracy against the primary source. Figures are drawn directly from the PIIE analysis cited above. Adoption statistics describe how many firms report using AI and should not be read as direct measures of job displacement.