AI is no longer arriving as a lab breakthrough or a boardroom slogan; it is arriving as a hiring-language shock. Nicholas Bloom at Stanford and colleagues have published a peer-reviewed article showing that mentions of AI in U.S. job postings appear to be roughly doubling each year since ChatGPT’s public debut, rising about twice as fast as cloud and four times as fast as smartphones.
That is a striking claim, and it fits the deeper logic of Bloom and his coauthors’ paper. When companies start rewriting job ads, they are no longer experimenting at the edge. They are budgeting, reorganizing, and betting on what work will look like next.
That is why this new evidence matters more than another headline about model launches or venture capital rounds. Bloom’s team built a map of technological diffusion using patents, earnings calls, and roughly 200 million U.S. job postings. Meanwhile, the 2025 Stanford AI Index shows that U.S. postings citing generative AI skills increased by more than a factor of three in 2024 alone. Together, those signals suggest that AI is moving from novelty to labor-market infrastructure at exceptional speed.
Why job postings are a reliable signal of AI demand
The Bloom paper is powerful because it treats vacancy text as evidence of actual demand rather than corporate theater. In the authors’ audit, 91% of technology mentions in job postings referred to tasks the worker would really perform, not generic brand language or boilerplate. That makes the hiring signal unusually clean.
The same study also finds that the technologies dominating executive discussion usually dominate labor demand too: just 276 technologies mentioned in more than 100 earnings calls accounted for about 39 million job postings, or 77% of all postings mentioning any new technology.
AI now looks like the newest and fastest member of that club. The Stanford AI Index, drawing on Lightcast data, shows that generative AI, large language modeling, and prompt engineering all surged sharply in job ads in 2024. This is not a story about a few Silicon Valley firms hiring model engineers. It is evidence that employers across the economy are starting to spell out AI capability as a concrete business requirement. The World Economic Forum’s Future of Jobs Report 2025 puts a broader frame around that: 86% of employers say AI and information processing technologies will transform their businesses by 2030. Job postings suggest many firms are not waiting for 2030.
That matters because vacancy data often catches change before the traditional statistics do. Productivity data arrive slowly. Capital spending is lumpy. Executive rhetoric is cheap. But a company that starts paying a premium for AI-literate workers is already changing how work gets done. The labor market is often the first place a technological revolution stops being theoretical. Speed, however, is not the same thing as breadth.
Is AI hiring spreading evenly across the U.S.?
One of the most important findings in Bloom’s paper is that major technologies diffuse through the economy far more slowly and unevenly than the hype cycle suggests. The authors find that the top five metro areas account for 33.3% of patents mentioning a new technology, and 42.1% for the most economically impactful technologies.
In the working-paper version summarized by the National Bureau of Economic Research, 56% of the most economically important technologies originate in just two places, Silicon Valley and the Northeast Corridor.
The geographic story is just as stark on the employment side. Bloom and his coauthors estimate that technology job postings can take decades to disperse fully across the country, with concentration falling only gradually over time. Even after 30 years, the average new technology remains far from evenly distributed. The Stanford AI Index shows the same pattern in miniature for AI today: California accounted for 15.7% of all U.S. AI job postings in 2024, followed by Texas at 8.8% and New York at 5.8%.
The skill pattern is equally revealing. In the year a technology emerges, 57.1% of related job postings require a college degree. Thirty years later, that figure is still 50.2%. New technologies broaden, but they stay skill-biased for a very long time. That should temper loose talk about instant democratization. Yes, AI tools are widely accessible. But the best jobs tied to a new general-purpose technology still cluster around the firms, cities, and workers that get there first.
There is one caveat, and it is an important one. The same Bloom research finds that lower-skill “use” jobs spread much faster than research, development, and production jobs. That means AI will likely travel outward in layers. The frontier work, model building, orchestration, and system redesign will stay concentrated. The applied work will spread later and more widely.
Can employers redesign work fast enough to capture AI gains?
That leads to the real question. The challenge is no longer whether AI can diffuse. It clearly can. The challenge is whether firms can reorganize themselves fast enough to capture the gains. The latest McKinsey global survey on AI finds that 78% of organizations now use AI in at least one business function, yet only 21% of those using generative AI say they have fundamentally redesigned at least some workflows. That gap is the whole story. Most companies are adding AI to existing routines when the larger payoff comes from rebuilding the routines themselves.
That finding aligns with the older NBER work on the productivity J-curve, which argues that general-purpose technologies create value only after firms make costly complementary investments in processes, skills, and organization.
It also aligns with newer evidence. In the NBER paper Generative AI at Work, access to an AI assistant raised customer-support productivity by 14% on average, with the biggest gains going to less experienced workers. And the PwC 2025 AI Jobs Barometer finds that industries most exposed to AI have seen 3x higher growth in revenue per employee, a 56% wage premium for workers with AI skills, and skill requirements changing 66% faster in AI-exposed roles.
What the data means for employers and workforce planning
Those numbers point to a simple conclusion. AI is not waiting for institutions to feel ready. It is already reshaping hiring, wage premiums, workflow design, and regional competition. The firms that treat AI as a staffing keyword will fall behind the firms that treat it as an operating model.
Bloom’s new job-posting evidence matters because it captures the moment when a technology stops being famous and starts being hired for. AI appears to be diffusing through labor demand at a historic pace, faster than cloud and far faster than smartphones.
But the broader lesson from the diffusion literature is less glamorous and more important: fast takeoff does not guarantee broad prosperity. The early winners will be the companies that redesign work, the regions that build talent pipelines, and the workers who learn how to use AI as a production tool rather than a parlor trick.
Everyone else will still be arguing about the future after the labor market has already moved on.
