The American economy is doing something unfamiliar: growing without hiring. Output expanded 2.2 percent in 2025 while payrolls added an average of just 15,000 jobs a month, a rounding error against a workforce of 160 million. Conventional models would call this stagflationary prelude. The data say otherwise. Productivity has run at 2.5 to 3 percent since late 2023, matching the best stretches of the postwar era. The interesting question is no longer whether the divergence is real, but what is causing it — and how long the trade lasts.
Three explanations compete, and they imply very different portfolios. The first, advanced by Citrini Research, attributes the productivity jump to AI adoption itself: capital is substituting for labour fast enough that aggregate output rises even as headcount falls. The second, from Noah Smith, is narrower — the productivity gain is largely a construction story, with data-centre buildout and computing equipment contributing to GDP at a pace the St. Louis Fed compares to the dot-com boom. The third, set out by The Economist, downplays AI almost entirely and credits broader technology diffusion across services, shale-era energy gains, and the labour-market reallocation that followed the pandemic.
Growing without hiring is either a productivity miracle or a measurement artefact. The trade is the same either way until it isn't.
The labour-market counterweight
The jobs picture complicates every version of the story. US tech employment is down 90,000 from its 2023 peak and has fallen by more than 10,000 in the past year alone — what Joseph Politano describes as the deepest sustained drawdown in tech employment since the dot-com bust, worse than either 2008 or the COVID shock. California, the historical centre of gravity, has watched its share of US tech jobs slide from nearly 19 percent pre-pandemic to roughly 16 percent today. Either AI is displacing the very workers who built it, or the sector is simply digesting a hiring binge. The two readings carry opposite implications for wage growth, consumption, and the Fed's reaction function.
The San Francisco Fed offers a more cautious read of the same numbers. Adjusted for capital utilisation, total factor productivity shows a moderate burst across 2023 and 2024 that faded in 2025 — closer to a one-off level shift than a new trend. If that revision holds, the productivity dividend currently being priced into long-run GDP forecasts is overstated, and the 2026 disinflation path that markets are leaning on becomes less reliable. Jerome Powell's remark that he never expected to see this many years of high productivity is best read as institutional surprise, not endorsement of a structural break.
Layered on top is a fiscal and liquidity impulse that flatters everything beneath it. Pre-midterm pro-growth measures and Fed balance-sheet expansion are easing funding strains across cyclicals, which is why Citrini's tactical book points to oil, tankers, and selective China exposure rather than to the AI names that ostensibly drive the thesis. Tyler Cowen's caveat is the sharpest: governments are poor judges of who actually works well with AI, so fiscal policy aimed at the transition will misallocate. The operational question for the next two quarters is whether productivity sustains above 2 percent once the data-centre capex cycle peaks. If it does, the soft-landing trade has another leg. If it doesn't, the jobs data stop being a curiosity and start being the signal.
Briefings are synthesised by the Ledger Desk from multiple sources cited in the sidebar. They are distinct from Articles, which are written by named contributors and carry a tracked Calibration Index. The Desk does not currently carry a Brier score; this is a deliberate choice for the v0.1 editorial layer and will be revisited.