Is AI an engine or a killer? Morgan Stanley provides answers with data: productivity soars, but nearly 90% of the benefits are taken by 12 industries

Zhihui Finance APP has noted: Is AI truly a productivity engine, or a job killer? A recent report by Morgan Stanley provides some answers with data: productivity is indeed soaring, but the benefits are highly concentrated; more crucially, employment in high-AI industries has started to “shrink.”

Nearly 90% of the benefits are captured by “leading industries”

Morgan Stanley used “output per employee” as a proxy for productivity and estimates that by the second quarter of 2026, overall US productivity will have risen by 2.4% year-on-year. Although this is a decline from 3.0% in the first quarter, it remains twice the average growth rate of the past twenty years.

It’s particularly notable that of this 2.4% increase, just 12 high-AI-exposure industries contributed 2.0 percentage points, while medium- and low-exposure industries combined contributed only 0.4 points. A year ago, high-AI industries contributed merely 1.4 percentage points—meaning concentration has only increased. The report bluntly states that productivity growth is “almost entirely driven by high-AI industries.”

Six months ago, Morgan Stanley attributed this round of productivity improvements to capital deepening from AI infrastructure investment, rather than business process reengineering. The current update broadly maintains that conclusion, but with a new finding: employment in high-AI industries has already begun to decrease, providing an additional boost to productivity. In other words, “machine replacing man” is showing early signs in the data—for now, however, the team characterizes it as a “micro, not macro” factor, “distant and secondary.”

Who is being replaced?

The report breaks down the employment changes line by line: data processing and software industries are accelerating layoffs even as output surges, and the credit intermediation sector (commercial banks and mortgage-related jobs) is also cutting staff more rapidly—these two areas are judged “highly likely to be AI-related.” Shrinking in private education and corporate management roles, on the other hand, is more likely due to tighter federal regulation, a decline in young populations, or aftershocks of last year’s upheaval, and may not be attributable to AI.

The computer manufacturing industry continues to hire, but both output and productivity have fallen; securities and miscellaneous professional services, by contrast, remain independently robust. Morgan Stanley concludes: while AI is causing job displacement in one or two high-AI industries, the main story behind strong productivity growth is still output expansion.

For the market, the value of this report is in turning the core question—“Is AI worth such expensive capital expenditure?”—into two quantifiable curves: on the one hand, investments in AI infrastructure are genuinely generating productivity dividends, lending macro support to the “AI narrative”; on the other hand, the substitution effect, albeit small, is growing. Coupled with previous research by the team on “pressures faced by young workers in high-exposure professions,” AI’s differentiation effect on the labor market is moving from theory to data.