The Data Came In. The Thesis Holds
Three weeks ago, we argued that the U.S. economy was producing a divergence that the headline GDP number obscures: output rising, jobs flat, productivity surging — and that AI-driven efficiency was the most plausible mechanism.
We said watch three indicators. This week, all three reported.
What the Productivity Report Showed
The Q4 2025 productivity data landed March 5. Nonfarm business productivity grew 2.8% annualized in Q4. Output rose 2.6%. Hours worked fell 0.2%.
One detail buried in the release deserves attention: the labor share of output — the percentage of what the economy produces that flows to workers as compensation — hit 53.8% in Q4 2025. That is the lowest reading since the Bureau of Labor Statistics began tracking the series in 1947. Capital captured a larger share of output than at any point in modern American economic history.
The Q3 surge in productivity (revised up to 5.2%) was dramatic enough to invite skepticism. Q4 at 2.8% looks more like the thesis settling into a durable trend than a one-quarter anomaly. Output growth with shrinking hours worked. Two quarters in a row.
What the Jobs Report Showed
The February employment report, released March 6, was bad — and the revisions made it worse.
Nonfarm payrolls fell by 92,000 in February. The market had expected a decline of around 50,000. But the headline understates the sequential deterioration: December, originally reported as a gain of 48,000, was revised down to a loss of 17,000. Combined with a smaller January revision, we entered February already 69,000 jobs lighter than previously reported.
Healthcare shed 28,000 jobs, the largest single drag — but this is largely explained by a strike at Kaiser Permanente that sidelined more than 30,000 workers. That component is expected to bounce back in March and shouldn't be read as structural. The more durable signals are elsewhere: information sector employment continued its steady decline, federal government payrolls fell another 10,000 (down 330,000, or 11%, since October 2024), and professional and business services — the knowledge worker category most exposed to AI substitution — showed no growth.
What the Third Indicator Showed
The third indicator we flagged was professional and business services employment. It covers the analysts, administrators, consultants, and knowledge workers in the salary band most vulnerable to AI displacement. In February, it was flat — unchanged, for practical purposes — even as the broader economy was contracting.
Flat isn't catastrophic. But flat, sustained over multiple months, while GDP holds and productivity rises, is exactly the pattern the thesis predicts.
Where This Leaves the Argument
The honest assessment: the data is consistent with the AI-productivity thesis, but it doesn't prove it. February's weakness had confounding factors — the Kaiser strike, ongoing federal workforce reductions under DOGE, and potential weather effects on construction. Analysts who want to dismiss the divergence have ammunition.
What they don't have is a counter-narrative that fits all the data. GDP above 2%. Productivity elevated two quarters running. Jobs contracting. Labor share at a 78-year low. Professional and business services flat. These don't collectively point toward a cyclical soft patch — they point toward something structural.
The divergence isn't closing. Next week, attention shifts to NVIDIA's GTC conference — where the companies building the infrastructure for AI productivity will be making the case for why this is only the beginning.
MasicotAI — Tracking the intersection of artificial intelligence and economic reality.