Why Technologies Fail to Spread

The ancient Greeks had a working steam engine. Hero of Alexandria described one in the first century AD — the aeolipile, a hollow sphere mounted on two tubes, which spun when water inside was heated and steam escaped through bent nozzles. It worked. Roman engineers could have seen it. Roman smiths had the metallurgical skill to build it in quantity. Roman society had capital, organized labor, and complex large-scale construction projects that could have benefited from mechanical power. None of this happened. The aeolipile remained a curiosity, displayed in temples as a demonstration of divine power, and the Roman economy continued to run on human and animal muscle for another fifteen centuries until European engineers, independently, worked out the same principle and used it to pump water out of mines.

This is not a story about ancient people being too ignorant to recognize a good idea. It is a story about the conditions that must exist before a technology does what it is capable of doing — and those conditions are almost never purely technical. The steam engine spread in eighteenth-century Britain because coal was expensive and mine drainage was an urgent problem and labor was scarce in the right places and there were investors willing to fund experimental machines and there was a legal framework to protect the inventors’ returns. Change any one of those variables significantly and the industrial revolution might have remained a thought experiment for another century.

The Adoption Gap Is the Normal Condition

Economic historians have spent considerable energy documenting what they call “technology adoption lags” — the gap between when a technology becomes available and when it becomes widespread. These lags are typically measured in decades. Sometimes in centuries.

The printing press reached Europe in the 1450s with Gutenberg’s movable type. But movable type had existed in China since the eleventh century, when Bi Sheng developed a system using fired clay characters. Chinese printing technology was not secret; it was documented and used. It produced books. And yet it did not produce the same cultural and economic transformation in China that the press produced in Europe four centuries later. The standard explanation — that Chinese characters are too numerous for movable type to be efficient — is partially true but insufficient. The deeper answer involves the structure of Chinese literacy, which was concentrated in a scholarly-bureaucratic class that had specific interests in controlling the production and distribution of texts, combined with a paper-making and woodblock printing industry that was already highly developed and highly profitable. Movable type arrived into an ecosystem that did not need it in the same way that European manuscript culture needed it.

The lesson is not that Chinese culture was backward or resistant to progress. The lesson is that technology adoption is not a simple function of technological superiority. It is a function of the fit between the technology’s economic profile and the existing economic and institutional environment. The printing press gave European merchants, scholars, and religious reformers something they desperately wanted: cheaper texts. The same technology offered Chinese printers something they already had: a way to make books. The different outcomes reflect different prior conditions, not different levels of sophistication.

The Price of Complementary Goods

A technology rarely fails to spread because it does not work. It fails to spread because the complementary goods and systems it requires do not exist, are too expensive, or are controlled by interests that benefit from the technology’s absence.

The electric car is an illuminating case, precisely because it is recent enough to document with precision. Electric vehicles were technically competitive with internal combustion vehicles in the early years of the twentieth century. The Detroit Electric, produced from 1907 to 1939, had a range adequate for most urban driving, required no hand-cranking, and was quieter and simpler to operate than the competing gasoline cars. For a brief period around 1910, electrics outsold gasoline vehicles in urban markets. Then the calculus shifted. Charles Kettering’s self-starter, introduced on the 1912 Cadillac, eliminated the dangerous and difficult hand-cranking that had been gasoline cars’ main disadvantage. The nationwide buildout of gasoline distribution infrastructure, driven by the oil industry’s substantial capital, made fuel available everywhere. The electric’s range limitation, acceptable in a city, became a decisive disadvantage in a country where drivers increasingly wanted intercity mobility.

The electric vehicle did not lose on technical grounds in any simple sense. It lost because the complementary infrastructure investment went to its competitor, and once that investment was made and the institutions built around it — the service station networks, the parts supply chains, the automotive culture of long-distance driving — reversing the choice required overcoming an enormous accumulated advantage. The technology that won was not necessarily better in any abstract engineering sense. It was better at the specific moment when the infrastructure question was being decided.

This pattern — complementary infrastructure determining which technology scales and which remains a niche — appears across every sector. Nuclear power in the United States faced not a technical barrier but an institutional one: the regulatory apparatus that had been built around coal and natural gas generation, the utility business models that depended on fuel supply relationships, the legal liability framework that made nuclear insurance prohibitively expensive. These were not natural facts about the world. They were accumulated institutional choices made by specific interest groups at specific historical moments. They could have been made differently.

When Incumbents Prevent Adoption

The interests that benefit from existing technology have both the motive and, usually, the means to slow the adoption of replacements. This is not conspiracy theory. It is straightforward industrial economics. A firm that has invested in a specific technology over decades has sunk costs in equipment, skills, supplier relationships, and regulatory expertise. New technology threatens all of that simultaneously. The rational response is to impede the new technology’s adoption — through lobbying, through standard-setting, through litigation, through the control of distribution channels — for as long as possible.

The gas lighting industry’s response to electric lighting in the 1880s is the canonical case. When Thomas Edison’s system of electric lighting was introduced, the gas companies did not simply compete on price and quality. They lobbied municipal governments to restrict electrical installations, spread stories about the fire hazard of electrical wiring, pressured insurance companies to charge higher rates for electrically lit buildings, and attempted to control the patents on incandescent bulb manufacturing. None of this prevented the eventual spread of electric lighting. But it slowed the transition by years, during which time the gas companies extracted rents from a technology they knew was obsolete.

The pharmaceutical industry’s management of drug patent systems provides a more recent parallel. A pharmaceutical company holding a profitable drug patent has every incentive to find ways to extend the effective patent life beyond the statutory term — through minor reformulations that qualify for new patents, through authorized generic agreements that delay generic entry, through risk management program requirements that create barriers to generic distribution. These strategies are legal. They are also straightforwardly rent-seeking behavior by an incumbent using institutional leverage to slow the adoption of cheaper alternatives. The technology (generic drugs) works. The institutional environment is shaped to reduce its spread.

Understanding this pattern matters because it changes how we should evaluate technology policy. The question is not usually whether a technology works. It is usually whether the institutional environment is organized to allow it to spread at the speed its technical merits would justify. The answer is frequently no, and the discrepancy is not accidental.

The Role of Labor Costs

One of the most durable findings in the history of technology adoption is that technologies spread fastest when they substitute for expensive things and slowest when they substitute for cheap things. This sounds obvious but its implications are counterintuitive.

The mechanization of textile production in Britain happened faster than in India not because British engineers were cleverer than Indian ones but because British textile labor was expensive relative to capital and Indian textile labor was not. The spinning jenny, the water frame, and the power loom were not solutions to a universal problem. They were solutions to the specific problem of expensive British labor. India’s abundant cheap labor made the same machines economically unattractive at the same price point. The result was that British textile production mechanized rapidly while Indian textile production continued with traditional methods — not because British society valued progress and Indian society did not, but because the relative prices of labor and capital were different.

This price-of-labor principle has predictive power across historical cases. The American economy mechanized agricultural production earlier and more thoroughly than European economies throughout the nineteenth and early twentieth centuries, and the reason was consistent: American agricultural labor was scarce and expensive relative to European agricultural labor. The labor scarcity was a product of geography (vast land relative to population), immigration patterns, and the expansion of industrial employment that competed for rural workers. The expensive labor created constant pressure to find mechanical substitutes, which funded the development and deployment of mechanized harvesters, planters, and processing equipment that European farmers, with access to cheaper labor, had less urgency to adopt.

This dynamic is not confined to history. Automation spreads fastest today in sectors and geographies where labor is expensive. It spreads slowest where labor remains cheap. The economic logic is exactly what David Hounshell documented for nineteenth-century American manufacturing: adoption follows relative prices, and relative prices differ across times and places in ways that profoundly affect which technologies get deployed and which sit in laboratories waiting for the economic conditions to catch up.

The Institutional Prerequisite

There is a final category of adoption barrier that deserves explicit treatment: the institutional prerequisite. Some technologies require institutional arrangements that do not yet exist before they can operate at scale, and creating those arrangements is often harder than developing the technology itself.

The railroad illustrates this with particular clarity. The engineering of steam locomotion on rails was achieved in the first decades of the nineteenth century. Trevithick’s locomotives worked. Stephenson’s Rocket worked. The physical technology was available. What was not available, initially, was the institutional infrastructure: the legal framework for expropriation of private land along rail routes (solved through specific parliamentary legislation in Britain), the corporate structures for assembling the capital needed to build rail lines over long distances (solved through the joint-stock company), the signaling and scheduling systems needed to prevent collisions on single-track lines (solved through decades of operational experiment), and the standardization of track gauge that allowed cars to move between different rail systems (solved — partially, slowly, and with enormous political conflict — through a combination of government mandate and market pressure).

Each of these institutional problems was harder to solve than the engineering problems that preceded it. Britain had working steam locomotives for a full decade before it had the legal and financial infrastructure to build a national rail network. The United States built its transcontinental railroad only after the Civil War created political conditions that made the land grants possible, and even then the enterprise was ridden with fraud and mismanagement that reflected institutional immaturity in handling projects of that scale and complexity.

The pattern repeats in contemporary technology deployment. The technical barriers to self-driving vehicles were largely overcome by the early 2020s. The institutional barriers — liability law, insurance frameworks, regulatory certification processes, urban infrastructure adaptation — remained largely unresolved years later. The technology sat, mostly working, while the institutions needed to deploy it at scale were slowly and painfully constructed. This was not regulatory failure in any simple sense. It was the normal process by which society builds the institutional infrastructure for a genuinely new class of technology, a process that has never been fast and shows no sign of becoming so.

The Adoption Lesson

Technology’s history is mostly a history of things that did not spread, at the speed and to the degree that their technical capabilities would have permitted, because the economic, institutional, and political conditions for their adoption were absent. The exceptional cases — the handful of technologies that swept across societies at high speed — are worth examining precisely because they reveal what conditions must align for rapid adoption.

Those conditions are: expensive alternatives that the technology clearly undercuts, complementary infrastructure that already exists or can be rapidly assembled, institutional frameworks that protect the returns to early adopters, and an absence of powerful incumbent interests with the means to impede adoption. These conditions align rarely and partially. When they all align at once, you get an industrial revolution. The rest of the time, you get Hero of Alexandria’s steam engine spinning cheerfully in a temple while the economy outside runs on muscle.

Recognizing this pattern should change how we think about the future. The technologies that will shape the next generation are not primarily the ones being invented now. They are the ones for which the economic, institutional, and political preconditions are being slowly assembled — sometimes deliberately, sometimes accidentally. Spotting that assembly process in progress, identifying which complementary systems are being put in place and which incumbent interests are being weakened, is a more reliable guide to technological futures than any assessment of technical capability in isolation.