Bolt a motor to a line shaft and you’ve still got a line shaft
In 1987, the economist Robert Solow needled the entire computer industry with one line:
“We see the computers everywhere but in the productivity statistics.”
Three years later, a Stanford economic historian named Paul David published a short paper with a long shadow: “The Dynamo and the Computer” (American Economic Review, May 1990). His answer to Solow, more or less: we’ve seen this movie before. The first run was electricity, and it took four decades to pay off.
The numbers are humbling. Edison’s first central power stations opened for business in New York and London in 1881. By 1899, eighteen years in, electric motors accounted for less than 5% of factory mechanical drive horsepower in the United States. Electric lighting had reached 3% of American homes. Factory electrification took until the early 1920s to hit the 50% diffusion mark, roughly forty years after the dynamo went commercial. David’s dry joke: an observer in 1900 could have complained that electric dynamos were to be seen “everywhere but in the productivity statistics!”
What took forty years
When factories first electrified, most did the obvious thing. They pulled the steam engine, bolted an electric motor in its place, and let it spin the same system of overhead line shafts and leather belts that had run the plant for decades. Historians call this “group drive.” The factory looked modern. The power bill said electric. The actual work stayed where the steam engine had put it: how the machines were arranged, how materials moved, how the building itself was designed.
The results were ugly. The old shafting stayed on the books as capacity while new motors got stacked on top, so the capital-output ratio went up and productivity gains stalled. Electricity, bolted onto a steam-era factory, performed like an expensive steam engine. Worse, the retrofit made the numbers sag before they improved, which handed skeptics a full generation of evidence that the dynamo was overhyped.
The payoff came from a different design entirely: “unit drive,” one motor per machine. Drop the overhead shafting and everything about the factory gets renegotiated. Buildings could be lighter and single-story. Machines could be arranged by the flow of materials instead of by their distance from the shaft. Wiring was modular, so one department could be reconfigured without shutting down the whole plant. Variable-speed motors improved quality. Workshops got brighter and safer. The old rotating belts had swirled dust and grease through the air and taken fingers with some regularity.
And then the statistics moved. David’s estimate: roughly half of the 5-percentage-point acceleration in US manufacturing total-factor-productivity growth from 1919 to 1929, versus the prior decade, is statistically accounted for by the growth in secondary electric motor capacity. The motor of 1925 looked a lot like the motor of 1895. The factory around it was unrecognizable.
The 2026 version of group drive
Picture a 30-person plumbing company. The owner buys twelve AI seats at $30 a head, call it $4,300 a year. Everyone gets a chatbot tab. And every process stays put: the dispatcher still transcribes phone calls into the ticketing system by hand, the office manager still keys invoices line by line, quotes still get built by copying last year’s quote and changing the numbers. Six months in: costs up $4,300, output flat. The owner concludes, reasonably, that AI doesn’t work for the trades.
That’s group drive. The conclusion is wrong; the install was. Those seats are motors bolted to line shafts. Paying for new capacity on top of an unchanged process is exactly how the dynamo-era factories made their own numbers look worse.
What unit drive looks like on a Tuesday
Same hypothetical company, different move: pick one workflow and rebuild it on the assumption the technology exists. Take quoting. Today a tech finishes a site visit, scribbles notes, and the office manager spends a chunk of every day turning those notes into quotes. Call it 10 hours a week at $35 an hour fully loaded, about $17,500 a year on assembly alone. Rebuilt: the tech talks into a phone for ninety seconds in the driveway, the system drafts the quote from the voice note, the photos, and the price book, and the office manager reviews and sends it the same afternoon instead of Thursday. Back-of-napkin, that recovers most of the $17,500, before counting jobs won because the quote landed while the homeowner was still deciding. Same technology as the chatbot tab. Different factory.
The excuse that expired
David is careful about one point that should sting a little: the forty-year lag was largely rational. The mills were sunk capital: multistory buildings engineered around water and steam, still serviceable, expensive to scrap. The big gains had to wait for old plants to depreciate, new ones to be built, and a generation of factory architects and electrical engineers to learn the new design from scratch.
Now run that excuse for a modern SMB. The line shafts today are habits and SOPs: the intake script, the quoting routine, the “how we’ve always done it” Word doc. A habit carries no book value. There’s no depreciation schedule to wait out, no building to scrap, and the people who know the new design are a phone call away rather than a generation out.
We keep arriving at this idea from different directions. Mark Cuban made the modern version of the argument, and we wrote it up in Cuban’s right. The AI bill is for the workflow you never rebuilt. We made the case from first principles in Why process beats raw capability, from the owner’s chair in You can just decide to be AI-native, and through the labor-market lens in The Handloom Weaver Was the Call-Center Rep of 1820. David’s paper is the receipt for all of it. Thirty-six years old, and it reads like a field report from this quarter.
The factory owners of 1900 had a forty-year excuse with a book value. A habit depreciates the day you decide it does.