#543 1920 · US manufacturing industry (per economist Paul David) · Manufacturing
Factories bought electricity for 40 years and got almost nothing for it, because they kept building their floors around the shape of the steam engine they'd replaced
the problem
Swapping in a powerful new technology delivers almost none of its potential if the surrounding system was designed around the old technology's constraints
background
19th-century factories ran on one central power source — a steam engine or waterwheel — turning a single large shaft that ran the length of the building, with a maze of belts and pulleys distributing power down to every machine on the floor. That system dictated the entire factory's physical layout: every machine had to sit close to the shaft it drew power from, workflow arranged around mechanical proximity to power rather than around the logical sequence of production.
Electric dynamos became commercially available in the 1880s, and the obvious first move for factory owners was the obvious one: replace the steam engine with a single central electric dynamo, keeping the exact same shaft-and-belt distribution system and the exact same factory layout it required. Economist Paul David later found that this substitution alone delivered almost no productivity gain for roughly four decades — electrification looked, on paper, like a technology that had failed to live up to its promise.
what everyone would do
The obvious first move for factory owners was straightforward substitution — swap the steam engine for a single central electric dynamo while keeping the exact same shaft-and-belt distribution system and factory layout the old power source required, since it was the least disruptive way to adopt the new technology.
what they saw
David saw, and the factory owners who eventually captured real gains discovered, that electricity's actual advantage wasn't being a better central power source at all — it was that power could be distributed in small units to each individual machine, meaning the entire logic of factory layout, built around mechanical proximity to one shared shaft, no longer had to apply. The technology's real value was invisible as long as it was used inside a system still designed around the old technology's constraints.
the move
The real productivity gain only arrived once factory owners abandoned the central-power-source model entirely and adopted 'unit drive': giving every individual machine its own small electric motor, eliminating the central shaft and its layout constraint completely. Freed from needing every machine physically close to a shared power source, factories could be redesigned around production workflow itself — lighter, more modular, often single-story buildings arranged around the logical sequence of manufacturing rather than mechanical proximity to power.
why it works
Giving every machine its own small electric motor eliminated the central shaft and belt system entirely, since no machine needed a mechanical connection to a shared power source anymore. Freed from needing every machine close to a shaft, factories could be laid out purely around the logical sequence of production rather than around power-distribution constraints, enabling lighter, more modular, often single-story buildings and the ability to run individual machines independently instead of the whole shaft together. Because the productivity gain came from redesigning the surrounding system to exploit unit drive's real properties, not from simply substituting dynamo for steam engine, the gains only appeared once enough factories had actually rebuilt around that new logic — which took roughly forty years because reworking a factory's physical layout is a far slower, more capital-intensive change than swapping out the power source alone.
the payoff
David found that widespread unit-drive adoption didn't reach roughly 50% until the 1920s, some 40 years after commercially available electric dynamos, and that measurable productivity acceleration from electrification only appeared once that threshold was crossed — with electrification then accounting for roughly half of all US manufacturing productivity growth during the 1920s, a delayed but eventually enormous payoff once factories redesigned around the technology's real advantage rather than its literal substitution for the old one.
where it breaks
This pattern applies specifically to technologies whose real advantage requires restructuring the surrounding system to exploit — a technology that's simply a drop-in, more efficient replacement for an existing component, with no distributional or structural implications, wouldn't show the same delay, since substitution alone would capture most of the benefit immediately. It also depends on organizations eventually recognizing that the initial substitution approach is capturing little value and being willing to bear the cost of a full structural redesign; an organization that never makes that leap stays stuck at the 'technology looks like it failed to deliver' stage indefinitely, exactly as many factories apparently did for decades. And the lesson generalizes only as far as this same underlying condition holds for other technologies — it isn't a guarantee that every new technology eventually pays off once restructured around, only a warning that judging a technology's promise from its performance inside an unchanged surrounding system risks badly underestimating its real value.
what came after
Paul David's 1990 paper 'The Dynamo and the Computer' became one of the most cited works in economics on technology-adoption lag, explicitly framed as a historical parallel to the 1980s-90s 'computer productivity paradox' and, more recently, cited by economists analyzing why AI adoption has so far produced smaller measured productivity gains than the technology's apparent capability would suggest — the recurring lesson being that the surrounding system, not just the tool, has to be rebuilt.
references
- [1]The Dynamo and the Computer: An Historical Perspective on the Modern Productivity ParadoxAmerican Economic Review (Paul David), 1990dklevine.com
- [2]The AI Moment? Possibilities, Productivity, and PolicyFederal Reserve Bank of San Francisco, 2026frbsf.org