A product manager in San Francisco opens her laptop at 7:30 AM. Before her first meeting, she's used AI to draft a competitive analysis, summarize fourteen Slack threads she missed overnight, and generate three variations of a launch email. She didn't learn these tools in a class. They just showed up in her workflow, one integration at a time, and now she moves at twice the speed she did eighteen months ago.
Eight hundred miles east, an HVAC technician in Louisville starts his day the same way he has for twenty years. Paper work orders. A phone call to dispatch. A mental map of the day's jobs built from experience, not algorithms. He's heard of ChatGPT. His kid uses it for homework. But nothing in his professional life has changed, and nobody's offering to change it.
These two people live in the same country. They will not live in the same economy much longer.
The Premium
Here's a number that should make you uncomfortable: workers with AI skills now command wage premiums up to 56% higher than their peers. That's not a PwC projection for 2030. That's now, in their 2025 Global Workforce Survey. Fifty-six percent. For knowing how to use tools that didn't exist three years ago.
Think about what that means in practice. Two software engineers, same company, same title, same years of experience. One learned to use AI-assisted development tools. The other didn't. The first one is worth half again more on the open market — not because she's smarter, but because she picked up a skill during a narrow window when it mattered.
And that window is already closing.
IDC projects that more than 90% of global enterprises will face critical AI skills shortages by 2026. Not "might face." Not "could experience." Will face. The price tag they put on the gap: $5.5 trillion in unrealized value. That's not lost revenue — it's the distance between companies that figure this out and companies that don't. Between workers who get pulled up and workers who get left where they are.
Meanwhile, McKinsey's 2025 State of AI report found that 88% of organizations say they're "using AI." Sounds great until you read the fine print: employees are three times more likely than their leaders expect to be using generative AI for more than 30% of their daily work. Translation: the executives think this is a controlled rollout. The workers know it's already everywhere. The gap isn't between companies that adopt and companies that don't. It's between leaders who understand what's happening on the ground and leaders who are reading last quarter's strategy deck.
Same Faces, Different Century
So who's actually building this stuff? Take a guess.
71% of AI talent are men. Only 26% of the global AI workforce are women, according to a joint analysis by Interface EU and LinkedIn. If you're over 55, your odds drop further: just 22% of baby boomers have received any AI training at all, per Randstad's 2024 workforce report.
Young. Male. Probably has a computer science degree from one of about thirty schools. If this profile sounds familiar, it should. It's the same demographic that built the internet, dominated early social media, and created the gig economy. Each of those revolutions promised to democratize something. Each one mostly enriched the people who were already closest to the power switch.
This isn't a conspiracy. It's a pipeline. The same universities that produce disproportionate tech talent are now producing disproportionate AI talent. The same companies that hired from those schools are now building AI labs. The same venture funds that backed Web 2.0 are now writing checks for foundation models. The system isn't broken. It's working exactly as designed — which is the problem.
When 74% of your AI workforce shares one gender and one generation, you're not building technology for humanity. You're building technology for a demographic. And that demographic's blind spots become the technology's blind spots. We've seen this movie. Facial recognition that doesn't work on dark skin. Credit algorithms that penalize women. Hiring tools that filter out anyone who doesn't look like the engineers who built them.
The new wave of AI isn't immune to these patterns. It's accelerating them.
Consider the feedback loop. AI systems learn from data. The people who choose what data matters, how to weight it, what "good" looks like in a training set — they're overwhelmingly from one slice of humanity. Not because of malice. Because of math. If your hiring pipeline pulls from thirty universities and those universities graduate 70% men, your AI team will be 70% men. If your AI team is 70% men, the products they build will reflect the problems they notice, the edge cases they catch, the use cases they imagine. The products ship. The market rewards them. The company hires more people like the ones who built the successful product. Rinse. Repeat. Calcify.
Breaking that loop requires more than diversity initiatives and scholarship programs — though those help. It requires acknowledging that the loop exists at all, which most of the industry hasn't done. The preferred narrative is meritocracy: the best people rise, the best tools win, the market sorts it out. Meritocracy is a comforting story for the people the system already selected.
Fifteen Cities and Everywhere Else
Geography makes the talent divide worse. A 2025 analysis by Second Talent found that 67% of global AI talent is concentrated in just fifteen cities. San Francisco, obviously. New York. London. Beijing. A handful of others. If you drew a circle around each one, you'd cover maybe 2% of the world's landmass and two-thirds of its AI capability.
This isn't just a stat for policy papers. It means that if you're a company in Nashville or a government office in Boise or a hospital system in rural Georgia, the people who could help you implement AI don't live within five hundred miles of you. They're in markets where the median rent is $3,000 a month, working for companies that can pay 56% wage premiums, building tools that may or may not solve problems you actually have.
Remote work was supposed to fix this. COVID proved people could work from anywhere. What it actually proved is that some people can work from anywhere — specifically, the people whose skills were already in demand. AI talent concentration hasn't decreased since the pandemic. It's increased. The cities that had the most AI workers in 2020 have more now. The places that had few have fewer.
There's a compounding effect here that doesn't get enough attention. AI talent clusters around AI talent. Engineers want to work with other engineers who are working on interesting problems. Companies build labs where the talent already is. Universities in those cities get more funding, produce more graduates, and feed more people into local companies. It's a flywheel, and once it's spinning, it pulls everything toward the center.
For the rest of the map — which is most of the map — the flywheel works in reverse. Fewer AI workers means fewer AI companies means fewer reasons for AI workers to move there means fewer AI workers. It's not a gap. It's a gravity well.
The consequences are already visible in hiring data. Companies outside major tech hubs report AI roles staying open two to three times longer than in coastal cities. When they do fill them, they pay relocation packages or remote premiums that strain budgets built for local cost-of-living. A hospital system in Mississippi competing with Google for machine learning engineers isn't a fair fight. It's barely even the same sport.
And here's what makes the geography problem different from the demographic one: you can theoretically diversify a workforce by changing hiring practices. You can't move San Francisco to rural America. The physical clustering of AI talent creates a knowledge density that remote work can supplement but not replace. The hallway conversations, the meetups, the ambient awareness of what's possible — these things have geographic coordinates, and those coordinates are expensive ones.
We've Done This Before
In 1995, 14% of American adults used the internet. By 2000, it was 50%. Five years to go from curiosity to necessity. That adoption curve reshaped every industry, created trillions in value, and left behind anyone who couldn't get online — which, for years, meant anyone who wasn't white, urban, and middle-class or above.
We spent the next two decades trying to close the digital divide. We built programs and funded libraries and subsidized broadband. Some of it worked. Most of it was too late. By the time everyone could get online, the people who got there first had already built the platforms, captured the network effects, and locked in the advantages. Getting a broadband connection in 2015 didn't put you on equal footing with someone who'd been building websites since 1997.
AI is compressing that same curve into about eighteen months.
Not five years. Eighteen months. The tools that barely existed in early 2023 are now embedded in enterprise software, creative workflows, legal research, medical diagnostics, and financial analysis. Companies that started experimenting in 2023 are now deploying at scale. Companies that are starting now are already behind. And individuals who haven't touched these tools yet aren't just late — they're watching the ladder get pulled up.
The internet parallel is instructive, but it understates the stakes. The internet changed how we access information. AI changes how we produce it. That's a different category of disruption. When the internet arrived, a journalist who couldn't use email was slower. When AI arrives, a journalist who can't use AI tools is competing against someone who produces five times the output at comparable quality. Slower is survivable. Five-to-one is not.
And the wage data already reflects this. That 56% premium isn't for people who build AI. It's for people who use it. The premium for people who build it is higher still. We're creating a three-tier workforce: people who build AI, people who use AI, and people who compete against AI users without it. The third tier is the biggest, and it's getting squeezed from both directions.
What This Actually Is
Every corporate earnings call and think-tank white paper frames this as a "skills gap." The language implies that the problem is a shortage of trained people and the solution is training more of them. Run a bootcamp. Fund a certificate program. Add an AI module to the MBA curriculum. Problem solved.
That framing is comfortable. It's also wrong.
This isn't a skills gap. It's an access gap wearing a skills gap's clothes.
Skills gaps are symmetrical: anyone can theoretically close them. This one isn't. A 58-year-old insurance adjuster in Dayton doesn't have the same access to AI training as a 26-year-old product designer in Austin. Not because she's less capable. Because the training doesn't exist in her world. Her company isn't offering it. Her professional network isn't talking about it. The tools aren't showing up in her workflow the way they showed up for the product manager in San Francisco. She would have to go looking for something she doesn't know she needs, in places she doesn't know exist, during hours she doesn't have.
That's not a skills problem. That's an infrastructure problem. And infrastructure problems don't solve themselves.
The "skills gap" framing also lets the right people off the hook. If the problem is that workers need training, the solution is on the workers. Pull yourself up. Learn to code. Take an online course. This is the bootstrap mythology applied to a structural problem — and it works about as well as it always does, which is to say it works fine for the people who were already going to make it and does nothing for everyone else.
The access gap is geographic. It's demographic. It's generational. It's built into the physical and social infrastructure of who gets exposed to new technology and who doesn't. Closing it requires more than curriculum. It requires rethinking who AI tools are built for, where they're deployed, and who gets to decide.
None of that is happening at the speed the gap is opening.
Here's what I want you to sit with: the AI revolution isn't coming. For a specific slice of the workforce, it already happened. For everyone else, the distance is growing every week. Not because they're behind — because the ground is moving under their feet while the people ahead are riding the same ground in the same direction, faster.
That 56% wage premium isn't a reward for brilliance. It's a tax on timing. And right now, timing is determined by zip code, employer, age, and gender more than it's determined by talent.
If you read this and you're on the right side of the wall, do one thing: bring someone with you. Teach a colleague. Share a tool. Send your weird uncle a walkthrough. The early internet got better when more people used it. AI will too. But only if "more people" means something beyond the same fifteen cities and the same thirty schools.
The wall is vertical. It's going up fast. And the view from one side is very different from the other.
This is Part 1 of The Vertical Wall, a five-part series examining who benefits from the AI revolution — and who doesn't. Next: Part 2: The Company Divide — why your employer matters more than your skills.