The Dynamo and the Computer
Writing in 1990, the economist Paul A. David set out to reassure colleagues who were baffled by a contradiction of their moment: a wave of innovation in microelectronics and communications was cresting, yet measured productivity growth had stalled. Robert Solow had captured the mood in a single line—“We see the computers everywhere but in the productivity statistics.” David's argument is that this so-called paradox looks far less puzzling once you read it against the economic history of earlier general-purpose technologies.
The dynamo as precedent
His chosen analogy is the electric dynamo. Like the computer, it was a general-purpose engine: a nodal element of a physically distributed network, bound up in a web of complementary technologies, and freighted with questions of standardization and compatibility. And like the computer, it took a remarkably long time to pay off. The first central power station opened in the early 1880s, but factory electrification did not reach even half of mechanical drive capacity until the early 1920s—roughly four decades later.
Why the gains were delayed
The lag, David argues, was not a failure of the technology but a feature of how such regimes change. It was simply unprofitable to scrap still-serviceable plants built around steam and water power, so the earliest electrified factories clustered in fast-growing industries that were building from scratch. Wider adoption had to wait on the physical depreciation of older plants, the obsolescence of cramped urban sites, and the capital-formation boom of the 1920s.
The transmission system mattered too. Retrofitting an old mill with “group drive”—electric motors turning existing line shafts—left the original belts and shafting in place as idle capacity, which raised the capital-output ratio and held down total factor productivity. The real payoff came with the “unit drive,” a motor on each machine. That allowed factories to be redesigned outright: single-story and lighter, with flexible layouts, cleaner and safer floors, and modular wiring that no longer forced a whole plant to shut down for a single change. Realizing those gains, though, depended on a slow, decentralized learning process and a cadre of engineers and architects who had to be trained by doing—diffusion that ran slower than would have been socially optimal.
Measurement and the eventual surge
Many of electricity's benefits—brighter and safer workplaces, faster transit, finer machine control—slipped past the conventional productivity indexes entirely. Even so, David shows that by the 1920s the numbers caught up: about half of the five-point acceleration in U.S. manufacturing productivity growth between the 1910s and the 1920s tracks the spread of secondary electric motor capacity, with capital-saving advances in continuous-process manufacturing adding still more.
The caution
David is careful not to push the parallel too far. Computers are not dynamos: human-machine interfaces are subtler, and information is not electric current. Information carries no neat marginal cost, resists conventional measurement, and can produce “overload,” a congestion effect with no easy equivalent in the power grid. A firm's information structures, unlike its machinery, do not physically wear out, which lends them a stubborn inertia. Yet these qualifications only reinforce his central claim—that genuinely new technologies face a long and costly period of adaptation before their promise registers in the statistics.
Why it still reads well
The essay's lasting move is to reframe the productivity paradox as an ordinary episode in the transition between technological regimes rather than an anomaly demanding special explanation. The payoff from a general-purpose technology arrives only once the organizations around it have been rebuilt to use it—an argument that has aged unusually gracefully through the internet and, now, the AI era.
Source
David, Paul A. “The Dynamo and the Computer: An Historical Perspective on the Modern Productivity Paradox.” American Economic Review 80, no. 2 (1990): 355–61.