The productivity J-curve is a concept formalized by Erik Brynjolfsson, Daniel Rock, and Chad Syverson in “The Productivity J-Curve: How Intangibles Complement General Purpose Technologies,” circulated as NBER Working Paper 25148 in 2018. It explains why a powerful new general-purpose technology can appear to lower productivity for years before it raises it.
The mechanism turns on intangible assets. When firms adopt a technology like AI, they must also invest heavily in things national accounts barely measure: new processes, new business models, retrained workers, and reorganized workflows. During this build-out, real resources go into producing intangible capital that does not yet show up as output, so measured productivity dips. Later, as those intangibles start generating returns, measured productivity rebounds and overshoots, tracing out a J shape over time. The authors found substantial J-curve effects for software in particular, and estimated that productivity-adjusting for intangibles, total factor productivity was meaningfully higher than official figures suggested.
For anyone evaluating AI investments, the J-curve is a useful corrective to impatience. A flat or even declining productivity number in the early years of adoption is consistent with the technology eventually paying off handsomely; it is the signature of an economy in the middle of building the complementary capital that makes the technology useful. The risk it highlights is mistaking the bottom of the J for failure and abandoning the effort just before the payoff.