Brynjolfsson, Rock & Syverson · MIT Initiative on the Digital Economy
The Productivity J-Curve
When firms adopt AI, measured productivity initially declines as resources flow into unmeasured intangible capital — redesigned processes, workforce training, data infrastructure. The payoff emerges later, often dramatically, as those intangible investments begin generating measurable output.
Based on "The Productivity J-Curve" (2021) & "The Rise of Industrial AI in America" (McElheran et al., 2025)
The J-Curve Trajectory of AI Productivity
Measured total factor productivity (TFP) relative to baseline after adoption of a general-purpose technology
Empirical Evidence: AI Adoption in U.S. Manufacturing
Comparative productivity trajectories from McElheran et al. (2025) — Census Bureau data, tens of thousands of firms
Phase 1
The Descent
Resources flow into unmeasured intangible capital: process redesign, data infrastructure, workforce retraining. Measured productivity falls as investment is consumed building assets that don't appear on the balance sheet.
Months 0–18
Phase 2
The Inflection
Intangible investments begin generating measurable output. Workers have learned new tools, data pipelines produce reliable predictions, and processes stabilize. The critical strategy decision — harvest or reinvest — must be made here.
Months 12–36
Phase 3
The Acceleration
Compounding returns engage as each prediction capability enables new decisions, generating new data that further improves predictions. Early adopters outperform non-adopters in both productivity and market share.
Months 24–60+
−1.3pp
Average short-term productivity decline after AI adoption
4%
Of companies creating substantial value from AI (BCG 2024)
3.5
Avg. use cases for AI leaders vs. 6.1 for underperformers
4yr
Window before AI adopters outperformed non-adopters
Sources: Brynjolfsson, Rock & Syverson, "The Productivity J-Curve" (AEJ: Macroeconomics, 2021); McElheran, Li, Brynjolfsson et al., "The Rise of Industrial AI in America" (MIT IDE / Census Bureau, 2025); Boston Consulting Group, "Where's the Value in AI?" (2024). Chart is a conceptual illustration of the J-curve dynamics described in these studies, not a direct reproduction of their figures. Phase timelines are estimates from the strategy document "Prediction Machines, Creative Destruction, and the Productivity J-Curve."