As tempting as it is to credit artificial intelligence here, it would likely be premature.
U.S. nonfarm productivity grew a robust 2.8% in 2025, continuing a string of improved productivity compared to the pre-pandemic decade, where growth averaged just 1.2%.
Productivity, in simple terms, measures how much output the economy generates per hour worked. It rises when workers become more efficient, when firms equip them with better capital or when resources shift toward higher-value activities. All three appear to be at work today, but as tempting as it is to credit artificial intelligence here, it would likely be premature.
The first reason is timing. General-purpose technologies typically do not lift measured productivity immediately. Instead, they follow a J-curve, as firms spend heavily upfront to reorganize around the technology while the payoff arrives later. The costs of retraining workers, redesigning workflows, structuring proprietary data and rethinking management processes are all largely expensed rather than capitalized and thus not captured as output in GDP.
During this transition period, productivity gains often emerge first in tasks automated or accelerated by the technology. Those gains can translate into more productive firms (like a native AI company that can increase coding output 10x) but aggregate productivity only rises once a critical mass of businesses has rebuilt workflows, redesigned processes, reorganized labor and had enough time for those gains to outweigh the upfront costs. Survey evidence suggests firms remain much closer to experimentation than scaled AI transformation.
The second reason is measurement. While capital investment has surged in this AI race, roughly two-thirds of data center costs flow to imported components, which offset much of the investment boost to GDP. Meanwhile, many of AI’s benefits accrue directly to consumers in ways that official statistics fail to capture. When someone uses an LLM for travel planning, writing help or legal guidance, AI may have been enormously useful to them, but beyond a potential subscription fee, it didn’t raise spending. For the businesses providing those services, the payoff may eventually materialize in retention or pricing power, but that value remains invisible in productivity statistics today.
If AI is not yet the main explanation, several other forces could be:
- Pandemic restructuring: The pandemic reshuffled labor and business formation in ways that may have raised average efficiency. Lower-productivity firms were disproportionately disrupted, digital processes were adopted quickly economy-wide and the “Great Resignation” likely improved job-matching.
- Capital deepening: U.S. firms had already been investing heavily in software, intellectual property and R&D before the current AI race. With modernized capital equipment and software, productivity gains can compound over time.
- Labor scarcity: A tight labor market, exacerbated by a sharp slowdown in immigration, pushed businesses to substitute capital for labor by investing in software and automation and demanding more output from existing workers.
None of this argues that AI is not already impacting the economy, but the official productivity statistics are probably the wrong place to look. Task- and firm-level evidence of AI’s productivity gains remains compelling, and firm-level adoption is broadening. Still, the history of general-purpose technologies suggests AI’s largest contribution to productivity is still ahead of us.
This leaves the U.S. economy in a favorable position, entering the AI era with an already improved productivity base after the post-GFC slump. If productivity growth supports earnings and helps contain labor costs, the market implications are also positive, painting a more durable backdrop for both equities and fixed income.