The Dividends of Technological Revolutions Have Always Been Late
Someone shared a reference list of nine books, ranging from Maddison's millennial statistics to Moore's paper. Read together, they tell the same story: making something 100 times cheaper has happened many times in history, but the dividends take decades to arrive—and rarely match the brochure.
The baseline ledger
Maddison's millennial statistics give a sobering starting point: world output per capita was about 467 international dollars in year 1 CE and 667 international dollars in 1820. Over eighteen centuries, growth was barely 40%—less than 0.02% per year. In the 180-odd years since, output per capita grew more than eightfold. Economic growth is not humanity's norm; it is a special case of the last century-plus, and it "arrived a thousand years late."
Three cost collapses
Gutenberg started printing around 1450. By 1500, Europe had printed some nine million books—in fifty years, matching the scale of a millennium of manuscript copying. A skilled scribe needed months to copy a single Bible. Book prices fell to a fraction of manuscripts.
Ford deployed the moving assembly line in October 1913. Chassis assembly dropped from 12.5 hours to 93 minutes, and the Model T's price fell from $850 to $260. Cheaper and higher-volume became the same thing for the first time.
Moore's 1965 paper, "Cramming more components onto integrated circuits," extrapolated to 65,000 components per chip by 1975, with doubling every year. A decade later he revised the pace to every two years. The widely quoted "doubling every 18 months" version was never Moore's words—that was another Intel engineer combining transistor counts with performance. The most-cited form of the prophecy is precisely the sentence the prophet never said.
The accountants' ledger reads differently
The brochure of invention and the ledger of economics have always been two different books. Crafts calculated the steam engine: from 1780 to 1830, British economy-wide productivity grew only about 0.3% per year; nearly a century after Watt, steam's contribution to labor productivity growth climbed to only about 0.4 percentage points per year. During the golden age of the Industrial Revolution, steam was statistically invisible—most growth came from textiles, organizational change, and other factors.
Fogel ran a counterfactual on railroads: had the 1890 United States lacked railways, canals and roads as substitutes would have cost GNP less than 5%, with most estimates between 1% and 4%. This work helped win him the 1993 Nobel Prize—not because railroads didn't matter, but because he demonstrated how "how much it matters" should be calculated.
Gordon's account of America is blunter: the special century was 1870–1970. Electricity, running water, the internal combustion engine, and medicine transformed households and cities; after 1970, growth fell to about a third of the prior rate, with information technology only briefly lifting the curve between 1994 and 2004. The dividend of invention does not arrive in the decade of invention—it comes after infrastructure, skills, and organization have all been rearranged.
Finance ran ahead of production
Two items in the ledger arrived early. The Dutch East India Company held humanity's first IPO in 1602, and the Amsterdam Stock Exchange opened alongside—150 years before the Industrial Revolution. The risks of oceanic trade were securitized first, with dividend yields fluctuating between 12% and 40% from 1679 to 1772.
Even earlier was silver. From 1500 to 1800, the Americas produced 80% of the world's silver; the mountain at Potosí alone shipped over 150,000 tons, minted into the first global currency, much of it flowing into the Ming dynasty's tax system. The first global supply chain ran two centuries before the steam engine.
One sentence to synthesize
The nine books converge on a pattern: costs collapse first, organizations reorder next, the ledger arrives last. The printing press cheapened books in 1450; literacy and the Reformation rewrote Europe in the following decades. The assembly line cheapened cars in 1913; America's golden growth was cashed in over two generations. The transistor began doubling in 1965; productivity only acknowledged it in 1994.
These accounting tools—counterfactuals, productivity attribution—are now being applied to AI. Some are already running Fogel-style calculations: how much slower would the economy be without large language models? History suggests the same answer: the immediate impact will be disappointingly small, the impact a decade out surprisingly large, and the biggest dividend will be booked to infrastructure not yet built.