US Productivity Growth & the AI Inflection
78 years of output-per-hour data from BLS via FRED (OPHNFB, PRS series; TFP from MFPPBS). Two prior acceleration eras — post-war industrialization and the IT boom — both hit 2.8% annual productivity growth. The question: does AI create a third?
The productivity thesis
US nonfarm productivity has gone through distinct regimes: a post-war boom (2.8%), stagflation (1.2%), the IT revolution (1.8% rising to 2.8% at peak), and a prolonged Great Slowdown (1.2%) since 2005. The 2023–2025 data shows early signs of reacceleration — but from COVID noise or structural change?
Institutional consensus: AI adds +0.5 to +1.5 percentage points of annual productivity growth by the early 2030s. Against a CBO baseline of ~1.5%, this implies 2.0–3.0% realized growth — returning to IT-boom levels. On a ~$35T economy, each +0.5pp = ~$175B in incremental annual output. Most forecasters agree: gains accrue where AI meets operational complexity in the physical world — manufacturing, logistics, energy, infrastructure, defense.
Productivity growth by era (1947–2025)
Annualized average output per hour growth, Nonfarm Business Sector. BLS via FRED.
Annual productivity growth (2020–2025)
Year-over-year output per hour growth. Note the 2023–2025 reacceleration.
Productivity growth, year-over-year (2019-Q1 – 2026-Q2)
Output per hour vs the same quarter a year ago — BLS's headline 4-quarter change. Smoother than the volatile QoQ-annualized series and the framing readers actually care about. Latest: Q1 2026 +2.9% (BLS preliminary, May 7 2026 release). Background bands color-code productivity eras (matching the “Productivity growth by era” chart above).
Beyond the headline: sectors & total factor productivity
The nonfarm-business headline hides a split: manufacturing productivity has persistently laggedthe aggregate — relevant to the thesis, since discrete manufacturing is where AI's constraint-collapse has the most room to run. Total factor productivity (output beyond what capital + labor inputs explain) is the cleanest read on technology's contribution.
Blue = nonfarm business (PRS85006091) · dashed = business (PRS84006091) · clay = manufacturing (PRS30006091). BLS via FRED, quarterly annualized rates.
TFP private business (MFPPBS, index → YoY) · TFP manufacturing (MPU4910063) · manufacturing labor productivity (MFGPROD, index → YoY). Manufacturing labor productivity has been roughly flat-to-negative since 2010 — the sector AI has not yet reached.
AI productivity forecasts — institutional consensus range
Eight forecasts from academia, government, and industry. Sorted from conservative to aggressive. The quantifiable consensus: +0.5 to +1.5 percentage points of additional annual productivity growth.
Connecting productivity to concentration
The IT Boom (1996–2004) produced 2.8% productivity growth AND the highest-ever concentration surge in software markets (S1 rose from ~30% to ~48% composite CR3 in that period). The same pattern played out in internet markets from 2000–2010 as Google, Amazon, and Facebook aggregated their respective categories.
The thesis:AI-driven productivity gains in the 2025–2035 window will be concentrated in cognitive services (legal, accounting, consulting, insurance, IT services) and physical services (trades, facilities, waste, pest control) — the categories with the lowest rev/employee and highest SG&A ratios today. As AI compresses the labor and distribution constraints that kept these industries fragmented, concentration should follow the same trajectory software and internet markets experienced during the IT boom.
See the Concentration Over Time chart (Overview → enable projections) and the Labor & AI Impact analysis for the market-level evidence.
Source Quality Hierarchy
Every market has a primary source plus 1–3 validation sources. Revenue numerators come from SEC 10-K filings; denominators validated against the Economic Census where available.
Data vintage: Q2 2025 research compilation. Labels updated 2026-Q2. Tier 1 sources refreshed from FY2025 SEC filings where available. For full methodology see the Methodology page.