Every Nation That Dominated One Tech Era Got Left Behind in the Next

Britain invented the steam engine, built the first railway network, dominated global manufacturing for 70 years, and then lost the next technological era almost completely. By 1913, the United States produced more steel than Britain and Germany combined. Germany owned the chemical industry. The electrical revolution — Edison, Siemens, Westinghouse — happened without Britain playing any significant role.

This was not an accident. It was a structural trap, and it has repeated with eerie precision across every major technological transition since.

The argument I want to make is this: incumbency in one technology era systematically produces the conditions for failure in the next. The mechanisms are identifiable. They recur. And everyone involved in the AI geopolitics conversation — which is basically everyone in tech policy right now — is conspicuously uninterested in examining them.

The British case is the clearest because the fall was so complete and so fast. Britain’s textile and iron industries were globally dominant in the 1840s and 1850s. The machinery was expensive. The infrastructure — canals, railways built for steam — was enormous. The workforce was trained for those systems. The financial structures, the bank financing models, the export economy: all of it was organized around the incumbent technology.

When coal-fired steam gave way to electricity and oil, adapting required writing off enormous amounts of existing capital, retraining enormous portions of the workforce, and restructuring the financing systems. None of these things were technically impossible. Germany managed them. The United States managed them. Britain, the richest nation on earth in 1870, found them politically impossible. Every constituency that had capital invested in the existing system had rational incentive to slow the transition.

The economic historian David Landes documented this meticulously in his 1969 book The Unbound Prometheus. British entrepreneurs, he argued, were not less intelligent than their German or American counterparts. They were responding rationally to local incentive structures that happened to optimize for preserving the old order. The word for this in modern terminology is “local optima.” The entire nation was stuck in one.

The British government tried to respond. There were parliamentary committees, royal commissions, educational reforms. The Technical Instruction Act of 1889 was explicitly modeled on German technical education. None of it worked, because the problem wasn’t educational — it was structural. The capital, the talent, the institutional prestige, and the political power were all concentrated in industries that had no interest in being disrupted. Reform required fighting those constituencies. Britain mostly couldn’t.

The American story in computing is more complicated, and more instructive for today’s situation.

The United States won the first computing era. IBM dominated mainframes from the 1960s through the 1980s, with a market share that makes Google’s current search dominance look modest. When the personal computer revolution hit in the late 1970s, IBM should have lost — its entire organizational structure was calibrated for centralized, enterprise-priced hardware sold to IT departments on long sales cycles. PCs were cheap, sold through retail, and disrupted the sales model entirely.

What saved the United States wasn’t IBM. It was that the US had an ecosystem — MIT, Stanford, Berkeley, Caltech, UIUC, Xerox PARC, DARPA-funded research, pools of venture capital in Boston and the Bay Area, and a culture of garage startups — that was structurally capable of spawning the next thing independently of what IBM did. Apple, Microsoft, Intel, and dozens of other companies emerged from outside the incumbent structure. IBM itself eventually made the PC, famously using off-the-shelf components and an open architecture, which turned out to be a decision that benefited Microsoft and Intel far more than IBM. But the US ecosystem survived the transition because it wasn’t solely dependent on IBM.

Meanwhile, Japan was running the most ambitious state-directed technology initiative in history. The Fifth Generation Computer Project launched in 1982 with ¥50 billion in government funding and an explicit goal of leapfrogging the United States in computing by 1992. It failed completely. Japan, which was the dominant consumer electronics and manufacturing power of the early 1980s, bet its competitive advantage on the wrong next paradigm — logic programming, reasoning systems — and lost. By the mid-1990s, Japan’s electronics industry was in long-term structural decline.

The pattern holds at the company level too, with a consistency that should make anyone running a large tech company uncomfortable.

Digital Equipment Corporation dominated minicomputers in the 1970s and 1980s. At its peak, DEC was the second-largest computer company in the world. Ken Olsen, its founder, reportedly told a 1977 convention that he couldn’t imagine any reason why anyone would want a computer in their home. This is often cited as a famous failure of vision, and it is, but it’s a superficial reading. Olsen’s real failure wasn’t lack of imagination — DEC had engineers who could imagine personal computers. The failure was organizational. DEC’s entire sales force, its support structure, its pricing model, its partnership ecosystem: all of it was built for selling to IT departments, not to individuals. PCs required a different company, not better imagination inside the existing one.

DEC was acquired by Compaq in 1998 for $9.6 billion — which sounds like a lot until you note that DEC’s peak valuation in the late 1980s, in 1987 dollars, was around $25 billion.

Nokia is a more recent and more studied case. Nokia dominated mobile phones from roughly 1998 through 2007, with market shares above 40% globally. It had internal smartphone projects. Its engineers knew touchscreens were coming. The company had research programs that were, in some cases, ahead of what Apple eventually shipped. None of that mattered when the transition came, because Nokia’s organizational structure — built around hardware-first, operator relationships, and a culture that treated software as secondary — was the wrong structure for the smartphone era. Apple arrived from outside the mobile industry with no legacy structure to defend.

The mechanisms are worth naming explicitly, because they’re not mysterious once you see them.

First: successful incumbents develop deeply optimized organizations. Their processes, their incentive structures, their promotion criteria, their internal politics — all of it is calibrated for the existing market. These organizations are very good at executing in the existing paradigm. They are constitutionally bad at doing things that would cannibalize what they’re already selling, or that require fundamentally different capabilities than the ones that made them successful.

Second: the skills that made you dominant in one era are often wrong for the next. The people who understood steam engineering were not the right people to understand electrical engineering. The people who understood mainframe sales cycles were not the right people to understand software startups. You can hire new people, but you can’t immediately change the culture that assigns status, resources, and advancement to the skills that won the last battle.

Third: infrastructure is deeply sticky. Britain’s rail network was built for coal-fired steam. Its factory equipment was built for coal. Its port infrastructure, its urban geography, its financial systems for commodity trading — all of it was calibrated for a world where steam was the primary power source. Retrofitting for electricity was enormously expensive, and the political economy of who would bear those costs was paralyzing. Capital that had been sunk into the old infrastructure created constituencies for protecting it.

Fourth — and this is the one that gets missed most often — dominant nations and companies attract the talent of the previous era. Britain had the world’s best steam engineers in 1870. That was a liability in 1890. The best talent in your field converges on the dominant institutions and works on the dominant problems. When the dominant problems change, the talent pipeline takes a generation to redirect.

Now apply this to AI and you start to see why the current geopolitical framing is so unsatisfying.

The United States dominates the current AI era by most measures that matter. OpenAI, Anthropic, Google DeepMind, Meta AI — the frontier model organizations are overwhelmingly American, drawing on decades of accumulated research culture, enormous pools of venture capital, and the world’s strongest universities in machine learning and computer science. The compute infrastructure, primarily the Nvidia supply chain, runs through American companies. The cloud platforms — AWS, Azure, Google Cloud — are American. The hiring market for world-class ML researchers is centered in American institutions.

China is formidable and closing gaps in specific domains. But the gap is real and documented. Huawei’s Ascend chips are not competitive with Nvidia’s H100s for frontier training. The training runs that produce GPT-4-class models require supply chains that Chinese firms don’t fully control, particularly in advanced semiconductors. Chinese labs produce impressive research and capable models, but the leading edge of capability is still predominantly in the US.

What neither camp wants to examine is the historical base rate on what dominant positions in technology eras actually mean for the next one. Britain was the dominant power in 1880. By 1920 it wasn’t, through no single catastrophic failure — just the accumulated drag of optimizing for a technology that was becoming less central to the global economy.

The scenario worth taking seriously is not “China overtakes the US in large language models.” LLMs might not be the defining technology of the next phase. Neuromorphic computing, quantum-accelerated learning, entirely new architectural approaches to AI — the defining technology of 2035 is not obviously the one that’s attracting the most capital in 2026. And if it isn’t, whoever dominates current LLM infrastructure might be in exactly Britain’s position in 1880 — enormously capable, enormously invested, and therefore enormously slow to adapt.

The EU is a different problem. The EU’s position in current AI is weak by every measure: no frontier models, limited compute infrastructure, a regulatory posture that makes rapid deployment difficult, and a political economy that makes large coordinated bets difficult to execute. The instinct in Brussels is to respond with industrial policy — GAIA-X, the AI Office, various national champion strategies. These might produce something. But they might also produce the Fifth Generation Computer Project: enormous expenditure, political direction, and a systematic failure to adapt when the foundational assumptions prove wrong.

There is a version of the EU’s position that is actually strategically interesting, though it requires admitting things European politicians don’t want to admit.

You can deliberately opt out of a technology era you’ve already lost and concentrate resources on specific applications or the following era. Switzerland didn’t try to beat the United States in social media. It built a world-class pharmaceutical and precision manufacturing sector instead, playing to its existing strengths in regulatory credibility and precision engineering. This requires the political will to admit you lost this round, which is a very hard thing to admit when your voters expect you to compete.

What the history actually suggests is that the countries most likely to lead the next era of AI — whatever form it takes — are not necessarily the current frontrunners. They might not be the current also-rans chasing the frontrunners either. They’re the ones building the ecosystem conditions — research culture, legal infrastructure, talent pipelines, risk capital — that can generate the next paradigm from unexpected directions.

The US had that in 1975. The question of whether it still has it in 2026 — given the consolidation of AI capability into a handful of very large, very well-funded organizations with enormous legacy structures to defend — is more open than the current confidence in American AI dominance implies.

The British establishment of 1880 was confident too.