The mechanism of the J-curve is counterintuitive. During the early adoption phase, organizations divert labor and capital from activities that would have produced measurable output under the old paradigm toward activities — training, process redesign, organizational restructuring — that produce intangible capital. The intangible capital does not appear in productivity statistics. The diverted resources do. The result is a period during which the economy appears to be investing enormous resources and getting nothing in return, when in fact it is building the infrastructure on which future gains depend.
The electric motor transition provides the cleanest historical validation. When American factories adopted electric motors in the 1890s, measured manufacturing productivity declined relative to its pre-electrification trend. Factories were spending enormous sums on equipment, wiring, and production reorganization while producing no more output than before. The invisible investment was taking the form of organizational knowledge adapted to the new technology's capabilities. The productivity gains that materialized beginning around 1920 were so large they transformed American manufacturing — but they had been invisible for three decades, hidden in the J-curve's dip.
The depth and duration of the dip depend on three factors: the magnitude of complementary investments required, the speed at which organizations can make those investments, and the accuracy with which productivity statistics capture the gains once they materialize. AI ranks high on all three dimensions — it demands massive complementary investment, it evolves faster than institutional learning, and its gains are disproportionately intangible and therefore measurement-resistant.
The framework explains a phenomenon that puzzled observers in early 2026: the widening gap between organizations reporting transformative AI results and organizations reporting modest improvements from identical tools. The J-curve framework predicted exactly this pattern. Organizations that had made complementary investments — process redesign, skill development, workflow restructuring — lived on the far side of the curve. Organizations that had deployed AI without the complementary investments remained in the dip. Same technology, different positions on the curve.
The J-curve formalization emerged from Brynjolfsson's collaboration with Daniel Rock (then MIT Sloan, later University of Pennsylvania) and Chad Syverson (University of Chicago). Their 2021 paper in the American Economic Journal: Macroeconomics, titled The Productivity J-Curve: How Intangibles Complement General Purpose Technologies, built a formal model showing that general-purpose technologies systematically produce underestimated productivity growth in their early years and overestimated growth in their later years, as intangible capital is first built and then harvested.
The empirical validation at the firm level came in 2025 when Kristina McElheran, Mu-Jeung Yang, Brynjolfsson, and John Kroff published their study using U.S. Census micro-data. They found a J-curve pattern at the micro-level: the average effects of AI deployment were negative in the short run, followed by growth along multiple dimensions over time. The short-run losses varied by firm age and strategy, but the pattern was unmistakable — confirming the framework at the granular scale where investment decisions are actually made.
The shape is predictive. Transformative technologies systematically trace a J through productivity statistics — dip first, rise later — with the curve's geometry determined by complementary investments.
The dip is investment, not failure. What looks like technological disappointment is the visible signature of invisible capital formation.
Micro and macro validation. The pattern appears at both the firm level (in Census data) and the aggregate level (in productivity statistics across GPT transitions).
Three determinants of curve geometry. Magnitude of complementary investments, speed of organizational adaptation, and accuracy of output measurement.
Divergence during the dip. Organizations making complementary investments pull away from those that don't, producing widening variance in outcomes from identical technology.