Where does U.S. productivity growth stand?
- Morgan Stanley measures sector productivity as “output / number of employees”: the four-quarter increase through the second quarter of 2026 was 2.4%, unchanged from the second quarter of 2025, below the 3.0% recorded in the first quarter of 2026, but about twice the average pace of the past 20 years.
- The firm estimates that real U.S. GDP grew 2.1% over the past four quarters, close to its estimate of potential growth. In its view, strong growth mainly reflects a supply-side boost from productivity improvements, rather than overheated demand.
Why are productivity gains becoming increasingly concentrated?
- Over the past four quarters, highly AI-exposed sectors accounted for 2.0 percentage points of the increase in output per person (a total increase of 2.4%); a year earlier, their contribution was 1.4 percentage points. The combined contribution of medium- and low-exposure sectors fell from 1.0 percentage points to 0.4 percentage points, leaving the sources of gains more concentrated.
- Data revisions have changed the attribution for an earlier period: Morgan Stanley had previously underestimated output and output-per-person growth in 2023–2024 by about 0.3–0.5 percentage points, but the latest upward revisions were concentrated mainly in sectors with medium AI exposure. The firm believes the rebound in 2023 came too early to be attributed to AI; the further strengthening of highly exposed sectors in 2025–2026 was the newer source of impetus.
Are productivity improvements driven mainly by output or by reducing employment?
- The firm’s view is that the main drivers remain faster output growth and capital deepening. Compared with its analysis six months ago, declining employment in highly AI-exposed sectors has also begun to contribute to gains in output per person, but labor substitution remains a secondary factor.
- Employment fell in about half of highly AI-exposed sectors. Data processing and software showed both rapid output growth and an accelerating decline in employment, which is more consistent with labor substitution. Credit intermediation also shed jobs, but weakness in commercial banking and mortgage employment may partly be related to a soft housing market.
- Counterevidence shows this is not a common employment pattern across all highly AI-exposed sectors: hiring accelerated in computer manufacturing, but both output and productivity declined; in securities, both output and employment grew. Meanwhile, over the past year, the unemployment rate edged down and the pace of job growth picked up slightly. Morgan Stanley concludes that the overall labor market can still absorb employment pressures in these sectors.
How much can these data tell us about what is due to AI?
- The report groups 64 industries into high-, medium- and low-exposure categories based on occupational AI exposure, then compares changes in output and employment across the groups. Thus, “high-AI sectors contributed 2.0 percentage points” refers to the statistical contribution of a group of industries; it does not mean that AI alone created 2.0 percentage points of productivity growth.
- The measure also has limitations: output / number of employees is a sector-level proxy, not the more suitable measure of output / hours worked. Changes in productivity and employment in highly exposed sectors can help track the AI transition, but the readings also reflect factors such as sector output, capital investment and employment.