Earlier this week, Microsoft announced quarterly profits of more than $36 billion. The same week, it offered buyouts to around 60,000 of its workers. It is pouring hundreds of billions into AI infrastructure across the industry, and it is cutting the humans to pay for it.
That contradiction is now the defining pattern of the AI economy. But the most important damage is not happening where most people are looking. It is not in the headlines about mass layoffs. It is in the quiet disappearance of the first rung of the ladder.
A new study from Denmark's central bank, Danmarks Nationalbank, has done what most aggregate employment statistics cannot. It linked survey data on AI adoption to monthly payroll registers and tracked individual companies against their own previous growth trajectories. The findings are uncomfortable. Firms that adopted AI in 2023 were about 9 percent below their expected employment trend by September 2025 compared to firms that did not. The gap widened slowly, then held.
But the aggregate number hides the real story. The decline was concentrated almost entirely in two places: small businesses and young graduates in their first jobs.
Among adopters with fewer than 100 employees, employment was roughly 16 percent below trend. Among larger firms, there was no meaningful decline at all. And occupations most exposed to generative AI—the kind of entry-level analytical, administrative, and research roles that graduates traditionally fill—sat about 27 percent below trend. Workers under 30 absorbed the overwhelming share.
This is not a story about robots replacing people overnight. It is a story about companies quietly deciding not to make their next hire. The Danish labour market makes this visible because it is flexible: firms can simply stop recruiting without firing anyone. Long-serving staff stay. The door for new entrants narrows.
The mechanism is structural, not malicious. For a small firm with tight financing, adopting AI has upfront costs: training, integration, software, workflow redesign. That money comes out of a finite budget. The easiest cost to defer is the graduate they were planning to hire in Q3. The work still needs doing, but the AI tool does enough of it that the headcount justification disappears.
The result is a slow, invisible squeeze. Unemployment doesn't spike. Overall employment looks fine. But the pipeline that turns education into experience is quietly blocked. And that matters because entry-level jobs are not just jobs. They are the mechanism by which a workforce is renewed. If every firm wants experience but no firm provides it, the system begins to consume its own future capacity.
Microsoft's numbers sit on top of this same logic at a much larger scale. The company is not primarily cutting because AI has replaced its engineers. It is cutting because capital is being redirected toward AI infrastructure and the market rewards the optics of efficiency. The human cost is real, but it is treated as a financing adjustment rather than a strategic consequence.
The deeper problem is that none of this shows up cleanly in the data. The Danish researchers are explicit: the effects are small enough in aggregate that without monthly registers, they would be invisible. The exposure indices used to classify which jobs are "AI-vulnerable" are crude. Two-thirds of a medical secretary's tasks may be automatable while a teacher's may be one-fifth, yet the index scores them almost identically. As the report notes, different language models score the same occupations anywhere from 2.7 percent to 51.5 percent exposed. We are making structural decisions on top of measurement tools that cannot yet see the structure.
The honest conclusion is not that AI is destroying all jobs. It is that AI is redistributing opportunity away from the people who need it most, through a mechanism that is almost perfectly designed to escape political and statistical attention. Firms save money. Profits rise. Share prices move. And somewhere, a graduate sends their two-hundredth application into a system that has quietly stopped answering.
The companies that will survive the next decade are not the ones that automate the fastest. They are the ones that understand a simple systems truth: you can defer the first hire, but you cannot defer the consequences of a workforce with no entry point. The ladder does not need to be pulled up. It needs to be maintained. And right now, almost nobody is paying for that maintenance.
Comments
New comment
The line, "That contradiction is now the defining pattern of the AI economy. But the most important damage is not happening where most people are looking. It is not in the headlines about mass layoffs. It is in the quiet disappearance of the first rung of the ladder, sends the shivers. Where will it all head. I recall as a youth deciding on what to study and the stress. My heart goes outto young people trying to juggle the AI outlook.
Imagine
It's wild how tough this will be to unwind. Google thought adding AI to search was the ultimate upgrade, but it's unraveling their whole business model. The AI starves Google of ad space, which starves websites of clicks, which starves authors of a place to publish, which starves the AI of the content it needs to reference. Eventually, Google destroys the ecosystem entirely, and nobody has a budget to advertise anyway. A perfect systems trap.
Pick a career from that...