Deep dive
The Great Hiring Freeze
Unemployment looks fine; the doors are closing anyway. Seven charts on how Big Tech learned to grow without hiring, where the payroll money went instead, and why recent graduates and job-seekers are the first to feel a labor market that froze rather than crashed.
By the headline numbers, the American labor market of the mid-2020s looks healthy: unemployment near 4 percent, layoffs unremarkable. But nearly every dial that measures doors opening — hires per job, postings that convert, the fortunes of new graduates, workers' confidence they could find something else — points the same direction. This is not a crash. It is a freeze. These seven charts trace it from Big Tech's income statements, through the capex that replaced recruiting budgets and the AI capabilities it bought, down to the people standing outside the door.
01
Big Tech’s Growth-without-Hiring Years
Growth stopped needing people
The combination that defines the era: revenue up, headcount flat or falling. It appeared in 6 of 20 Big Tech company-years from 2022 onward — after appearing exactly zero times in the 25 company-years from 2017 through 2021. The business model changed before the labor statistics did.
What it adds
Revenue growth with flat or falling headcount appeared in 6 of 20 Big Tech company-years from 2022 onward, after appearing in none of the 25 observations from 2017–2021.
Open chart, data, and methodology ↗The money that used to flow to recruiters went somewhere else entirely.
02
Is AI Spending Replacing Hiring?
The capex went to compute
Since 2021, four hyperscalers' trailing capital spending rose 324 percent, while U.S. software-development job postings fell 32 percent and the same companies' reported headcount rose just 3 percent. Divergence is not proof of substitution — but it is exactly the shape substitution would have.
What it adds
Since 2021, four hyperscalers’ trailing capex index rose 398%, while software-development postings moved -28% and their reported headcount moved +3%. The divergence does not prove substitution.
Open chart, data, and methodology ↗What that compute buys is now measured in hours of human work.
03
The Task Horizon
The machines' workday
METR measures the length of software task frontier AI can complete at 50 percent reliability: about six seconds in early 2019, about seventeen hours — more than two working days — by spring 2026. The horizon has doubled again and again, on a log scale that keeps needing new gridlines.
What it adds
The length of software work frontier AI can complete reliably is increasing at an exponential pace.
Open chart, data, and methodology ↗An economy buying machine-hours still advertises jobs. It just fills fewer of them.
04
The Beveridge curve remains outward-shifted
Vacancies without hires
Since 2023, the job-openings rate has run about half a percentage point above what its 2010s relationship with unemployment would predict. Postings stay up; conversions don't follow. The openings data and the hiring data are telling two different stories, and the gap itself is the story.
What it adds
Since 2023, the job-openings rate has averaged about 0.5 percentage points above its 2010s relationship with unemployment.
Open chart, data, and methodology ↗Look beneath the postings and the engine of the labor market has visibly slowed.
05
Hiring Is Not Job Growth
The churn freeze
In 2000, the private sector hired 26 workers per 100 jobs every quarter. By 2025 it hired 16.4 and separated 16.3 — enormous flows that net out to +0.1 jobs per 100. American labor-market churn has fallen by a third over a generation, and net growth now lives inside a rounding error.
What it adds
In 2025-Q1, the private sector recorded about 16.4 hires and 16.3 separations per 100 beginning-of-quarter jobs. Those large flows produced net job change of only +0.1 per 100 jobs.
Open chart, data, and methodology ↗A frozen market is survivable for the seated. It punishes the people trying to get in.
06
Graduate unemployment
The young pay first
For most of three decades, a fresh degree meant lower unemployment than the workforce at large. That flipped: recent college graduates now run 5.6 percent against 4.2 percent for all workers, and the inversion has widened through 2026. The entry-level rung is where the freeze bites hardest.
What it adds
Recent college graduates now have a higher unemployment rate than the workforce as a whole.
Open chart, data, and methodology ↗Ask workers how it feels, and they describe the freeze with surprising precision.
07
The Worker Anxiety See-Saw
How it feels from inside
The New York Fed asks workers two questions every month. Over the past year, perceived risk of losing your job barely moved — while confidence in finding another within three months fell almost six percentage points. Workers are not afraid of being fired. They are afraid the door won't reopen.
What it adds
From July 2025 to June 2026, perceived job-loss risk was essentially unchanged while confidence in finding another job within three months fell almost six percentage points.
Open chart, data, and methodology ↗That asymmetry — safe in your seat, stuck in your seat — is the freeze in one sentence.
The quick read
A freeze, not a crash
Every chart in this story can coexist with 4 percent unemployment, which is exactly why the moment is easy to misread. Hiring per job is at generational lows, postings convert worse than they have in a decade, the graduate premium has inverted, and workers' fear has shifted from losing jobs to finding them. Whether AI is the cause or merely the excuse, corporate budgets moved from payroll to compute, and the labor market's doors swing more slowly than at any time in the modern record. The question these charts will answer next: does the freeze thaw as the capex digests — or is growth without hiring simply what growth looks like now?
Each chart is linked to its own methodology, source list, latest observations, and downloadable chart image. Associations shown here are descriptive unless a chart explicitly states otherwise.