Is AI an engine or a killer? Damo used data to give the answer: productivity soared, but nearly 90% of the dividends were eaten up by 12 industries

The Zhitong Finance App noticed that AI is a productivity engine or an employment killer? Morgan Stanley's latest report uses a set of data to give the answer: productivity is indeed booming, but dividends are highly concentrated; more importantly, employment in high-AI industries has begun to “shrink.”

Nearly 90% of the dividends were eaten up by “leading industries”

Using “output/employee” as a proxy productivity indicator, as of the second quarter of 2026, the overall productivity rate in the US rose 2.4% year on year. Although down from 3.0% in the first quarter, it was still twice the average growth rate of the past 20 years.

It is worth noting that in this 2.4% increase, only 12 industries with high AI exposure contributed 2.0 percentage points, while the medium and low exposure industries contributed only 0.4 percentage points in total. A year ago, the contribution of the high AI industry was only 1.4 percentage points — concentration did not decline but increased. The report bluntly stated that productivity growth was “almost entirely driven by the high AI industry.”

Half a year ago, Damo determined that this round of productivity growth stemmed from capital deepening brought about by investment in AI infrastructure rather than corporate process re-engineering. This update largely continues that conclusion, but there is one more discovery: employment in the high AI industry has begun to decline and is making an additional contribution to productivity. In other words, “machine substitutes” are showing signs in the data for the first time — it's just that the team characterized them as “micro rather than macro,” which is a “remote and secondary factor.”

Who is being replaced?

The report breaks down employment changes step by step: the data processing and software industry is increasing output while layoffs are accelerating, and credit intermediaries (commercial banking and mortgage-related jobs) are also speeding up staff reductions. These two categories were judged to be “likely related to AI.” However, the contraction in private education and corporate management positions is more likely due to tightening federal regulations, shrinking young people, or the aftermath of last year's turmoil; it is not necessarily due to AI.

The computer manufacturing industry continues to recruit people at an accelerated pace, but both output and productivity have declined; the securities industry and miscellaneous professional services have remained independent and strong. Damo's summary: AI has indeed replaced one or two high-AI industries, but the main narrative of high productivity growth is still output expansion.

As far as the market is concerned, the value of this report is that it breaks down the spiritual question of “whether AI is worth such high capital expenses” into two quantifiable curves: on the one hand, AI infrastructure investment is actually being transformed into productivity dividends to provide macro-level support for the “AI narrative”; on the other hand, the substitution effect is small but increasing. Combined with the team's previous research on “high-exposure professional youth pressure,” AI's differentiation of the labor market is moving from theory to data.