Morgan Stanley: 83% of U.S. Productivity Growth Comes from 12 High AI-Exposure Industries
nashnova research
Morgan Stanley's latest research finds that 83% of US productivity growth over the past year came from just 12 high-AI-exposure industries, while the remaining 52 sectors barely moved. The AI dividend remains a few industries' turbocharger — not an economy-wide lift.
83% of growth from 19% of industries — what does that actually mean?
US output per worker grew 2.4% over the past year. Of that, 2.0 percentage points came from just 12 high-AI-exposure industries. This means → the other 52 industries combined contributed only 0.4 pp — near-stagnation.
In plain terms = imagine a 64-student classroom where only 12 students are pulling up the average; the other 52 are barely progressing.
A year ago the gap was narrower: the high-AI group contributed 1.4 pp, the rest 1.0 pp. The spread has since widened from 0.4 to 1.6 pp — concentration is accelerating, not converging.
Which industries are surging, which are falling behind?
Data processing and internet publishing saw productivity grow 27.0% year-on-year, still accelerating each quarter. Securities and investment grew 17.6%. Publishing (including software) and computer systems design each grew roughly 9%. These sectors are AI's builders or direct beneficiaries.
The industries where most Americans actually work tell a different story: hospitals up just 0.3%, construction flat, food services up 0.9%, food and beverage retail down 4.7%, trucking down 11.4%, general merchandise retail down 11.8%.
Put simply = the code-writing, deal-making sectors are being propelled by AI. The plate-carrying, truck-driving, shelf-stocking sectors are not only missing the lift — some are moving backward.
Is the high-AI camp itself splitting apart?
Computer manufacturing's output growth plunged from 13.0% in late 2025 to -1.1% in Q2 2026. The cause: the industry kept hiring while output fell — headcount outran workload.
Legal services are widely seen as a frontier for AI disruption, yet productivity actually fell 1.8% — output stalled while payrolls rose. This means → "high AI exposure" does not equal "productivity already delivered." There is a gap between exposure and execution.
Was the earlier productivity improvement actually old momentum?
Morgan Stanley notes that prior annual GDP revisions lifted 2023–2024 output per worker by 0.3 to 0.5 pp, but this "appeared too early to plausibly reflect AI's contribution."
The bank judges these revisions concentrated in mid-AI categories, more likely driven by pandemic-era cost-cutting. This means → the data previously cited as proof that "AI is already boosting productivity" was telling a different story.
The report argues that the further acceleration in 2025–2026 is the first signal of new momentum from high-AI industries — old tailwinds are fading, new ones are taking over, but the handoff zone is extremely narrow.
3.6% nonfarm productivity — what's hiding behind the headline?
The Bureau of Labor Statistics reported Q3 nonfarm productivity grew at an annualized 3.6%, above the 2.7% consensus, with unit labor costs declining 0.7%. AI optimists seized on this as validation.
Morgan Stanley's industry-level breakdown shows the number masks extreme unevenness: a handful of sectors' explosive gains pulled up the overall average.
In plain terms = the mean looks strong, but the median barely budged. The AI dividend remains "productivity for the few." Whether it can spread to hospitals, construction, and transport — the sectors that employ most workers — is still the unresolved central question.
市场有风险,内容仅供研究参考,不构成投资建议。
