Explainer
The Productivity J-Curve
Why transformative technologies make productivity look worse before making it better, and what this means for AI.
When a major new technology arrives, measured productivity often stagnates or declines for years, not because the technology doesn’t work, but because the massive complementary investments it requires are invisible to official statistics.
The Productivity Paradox
In 1987, economist Robert Solow made a famous observation about computers: you could see them everywhere except in the productivity statistics. Billions were being spent on IT, yet measured productivity growth remained stubbornly flat.
This wasn’t a new pattern. When electric motors first appeared in factories in the 1880s, it took roughly 40 years before electricity showed up in manufacturing productivity data. The same phenomenon is now playing out with artificial intelligence: massive investment, widespread adoption, but little apparent impact on aggregate productivity.
Erik Brynjolfsson, Daniel Rock, and Chad Syverson argue this isn’t a mystery. It’s the predictable result of how we measure productivity, and how general-purpose technologies actually transform economies.
~40 yrs
Electricity
Introduction to productivity surge
~25 yrs
Computers
Adoption to productivity boom
?
AI's lag
2020s to ???
The J-Curve Explained
The J-Curve describes the systematic error in how we measure productivity when a general-purpose technologyA transformative technology that reshapes entire economies: steam, electricity, computers, AI. Economists abbreviate this "GPT" (not to be confused with the AI model "Generative Pre-trained Transformer"). is adopted. Measured productivity initially drops below true productivity, then later overshoots it. The result is a J-shaped curve in the measurement error over time.
Conceptual illustration based on Brynjolfsson, Rock & Syverson (2021)
Click or hover on a phase to learn more
Why It Happens: The Intangible Investment Gap
General-purpose technologies don’t just plug in and produce value. They require enormous complementary intangible investments: reorganized workflows, retrained workers, redesigned processes, new management practices, and co-invented business models.
Here’s the accounting problem: national statistics treat these intangible investments as current expenses rather than capital investments. When a firm spends $10 million redesigning its operations around AI, that spending reduces measured GDP instead of being counted as an addition to the capital stock.
The result is a two-sided measurement error that creates the J-curve shape:
The Measurement Gap
When firms invest in reorganization, training, and new workflows, national accounts treat these as current expenses that reduce measured GDP — not as investments that build future capacity. The economy looks like it's stagnating when it's actually building up an enormous stock of unmeasured intangible capital.
The economy looks like it’s stagnating when it’s actually building up an enormous stock of unmeasured intangible capital.
Early on, measured productivity is understated because firms are diverting resources from producing measurable output to building unmeasured intangible capital. The economy is building future capacity, but the statistics say it’s stagnating.
Later, measured productivity is overstated because the accumulated intangible capital stock is producing measurable output, but isn’t counted as an input. The gains appear to come from nowhere.
Three Waves, One Pattern
The J-curve has played out at least twice before. Each time, observers declared the technology overhyped during the investment phase, only to see a productivity surge once complementary investments matured.
Investment Phase
Factories initially just swapped steam engines for electric motors, keeping the same layout. The real gains required complete factory redesign: individual motors on each machine, single-story layouts, assembly lines.
The Paradox
Total factor productivity was essentially flat from 1890 to 1920 despite widespread electrification.
The Payoff (~40 years after introduction)
U.S. manufacturing TFP grew roughly 50% between 1899 and 1929, with electricity-intensive industries leading.
Hidden Intangible Investments
What This Means for AI Today
If the J-curve framework is correct, the current period of apparently disappointing AI productivity gains is not evidence that AI doesn’t work. It’s evidence that we are in the investment phase of a new general-purpose technology cycle, and the complementary intangible investments haven’t yet matured.
There are reasons to think AI’s J-curve may be compressed compared to prior GPTs. Software-based reorganization can propagate faster than physical factory redesign. AI models improve continuously (unlike static machines), and digital workflows can be replicated at near-zero marginal cost.
But there are also reasons it could be extended. AI touches more job categories simultaneously than any prior GPT, requiring more widespread organizational change. Trust and regulatory frameworks are still forming. And the intangible investments required (in data quality, workflow redesign, and human-AI collaboration skills) may be deeper than anticipated.
For patience
Apparent lack of productivity impact is predicted by the model, not evidence against it. The investment phase is necessary groundwork.
For caution
The J-curve doesn't guarantee a happy ending. The electricity analogy took 40 years. Workers displaced during the trough can't wait decades.
For measurement
Current productivity statistics may systematically understate AI's actual contribution. The authors' IT calibration estimated a gap of 11-16%.
For firms
The factory-level lesson is clear: superficial adoption (swapping tools without redesigning workflows) captures almost none of the value.
The productivity J-curve resolves an apparent paradox: how can a technology be transformative yet invisible in the data? The answer is that transformation requires investment, and our statistics are blind to the most important kind.
Source
Brynjolfsson, E., Rock, D., & Syverson, C. (2021). The Productivity J-Curve: How Intangibles Complement General Purpose Technologies. American Economic Journal: Macroeconomics, 13(1), 333–372.