The Second Divergence

Why the countries ignoring AI today are making the same catastrophic mistake as the countries that ignored industrialization in the 1800s

Why the countries ignoring AI today are making the same catastrophic mistake as the countries that ignored industrialization in the 1800s

In 1820, the average income of a person in Western Europe was roughly twice that of a person in Asia or Africa. By 1975, it was twenty times higher. The gap between rich and poor nations that defines the modern world — the inequality that shapes migration, conflict, trade, and politics — was not ancient. It was not inevitable. It was manufactured, in the span of roughly 150 years, by a single technological transition that some countries rode and others missed.

That transition was industrialization. And the countries that missed it didn’t miss it because they were incapable or unintelligent. They missed it for reasons that were structural, political, and in many cases the direct result of choices made by their own elites. The consequences lasted not years, but centuries. Many are still being paid today.

We are standing at the beginning of a transition of comparable magnitude. Artificial intelligence is not merely a new technology. It is a general purpose technology — one that transforms productivity across every sector simultaneously, the way steam power and electricity did before it. And just as in the 1800s, some countries are riding this transition aggressively while others are watching, waiting, debating, or simply unaware of the scale of what is unfolding.

The question history is asking right now, with uncomfortable precision, is this: which countries are making the mistake that India, China, and the Ottoman Empire made in the 19th century — and will they understand it before the gap becomes permanent?

What Actually Happened in the 1800s

To understand the parallel, you need to understand what industrialization actually did — not the romantic version with steam trains and cotton mills, but the structural economic reality.

Industrialization was a productivity multiplier of a kind the world had never seen. A factory worker operating a power loom produced cloth at roughly 40 times the rate of a hand weaver. A steam-powered mill processed grain at speeds that made manual milling economically obsolete almost overnight. The nations that industrialized early — Britain first, then France, Germany, the United States, and Japan — didn’t just get richer. They fundamentally changed the ratio of output to labor input, which meant they could produce more with fewer people, accumulate capital faster, fund larger militaries, and expand their economic and political reach globally.

The nations that didn’t industrialize, or industrialized late, didn’t just fall behind in relative terms. They found themselves unable to compete economically or militarily with powers whose productive capacity was now orders of magnitude greater. The Qing Dynasty in China had the largest economy in the world in 1820. By 1900, it was being carved into spheres of influence by European powers whose industrial base had transformed them from comparable civilizations into dominant ones in the span of two generations.

The Ottoman Empire — sophisticated, cosmopolitan, administratively capable — watched the same process unfold and responded with reforms that were perpetually too slow, too partial, and too captured by internal political resistance to matter at the scale required. Egypt under Muhammad Ali came the closest to a genuine industrialization strategy in the non-Western world, and it was deliberately undermined by British economic pressure designed to prevent a competing industrial power from emerging in the region.

India, the jewel of the British Empire, was actively deindustrialized — its textile industry systematically destroyed to create a captive market for British manufactured goods. This was not an accident of history. It was policy. The result was that India entered the 20th century as an agrarian economy despite having been, two centuries earlier, one of the world’s leading producers of fine textiles.

The lesson buried in all of this is not simply that industrialization was good and missing it was bad. The lesson is that when a general purpose technology transforms the productivity frontier, the gap it creates between early adopters and late adopters is not linear — it is exponential. And it persists.

AI Is a General Purpose Technology — This Matters Enormously

Most technologies improve one thing. Better seeds improve agricultural yields. Better ships improve maritime trade. Better weapons improve military effectiveness. General purpose technologies are different. They improve the productivity of nearly everything simultaneously.

Steam power was a general purpose technology. It transformed manufacturing, transportation, agriculture, mining, and eventually communication. Electricity was a general purpose technology. It transformed every sector it touched, from industrial production to domestic life to information processing. The internet was a general purpose technology. It restructured commerce, communication, media, finance, and governance all at once.

AI is a general purpose technology of the same order. It is already transforming drug discovery, materials science, software development, legal research, financial analysis, logistics, agriculture, education, and military strategy simultaneously. It is not a tool for one sector. It is a productivity multiplier for the entire economy.

This is precisely why the parallel to industrialization is not metaphorical. It is structural. When a general purpose technology arrives, nations that adopt it early compound their advantages across every sector at once. Nations that adopt it late fall behind across every sector at once. The divergence is not contained to one industry. It is total.

The economists who study general purpose technologies use a specific term for what happens at the frontier of adoption: they call it the productivity surge. It is not immediate — there is typically a lag of one to two decades between the adoption of a general purpose technology and the measurable surge in productivity as institutions, skills, and infrastructure reorganize around it. The countries that are investing in AI infrastructure and education today are planting the seeds of a productivity surge in the 2030s. The countries that are not are ensuring they will be watching that surge from the outside.

The Geography of the Current Divide

The AI investment landscape today maps onto the world’s existing power structure with uncomfortable clarity, and then extends it.

The United States is investing at a scale that is difficult to overstate. Microsoft, Google, Amazon, and Meta alone have announced combined AI capital expenditure plans exceeding $300 billion for 2025. The US government is funding AI research through DARPA, the NSF, and a growing network of national AI research institutes. American universities are producing the majority of the world’s top AI researchers. The infrastructure — the data centers, the chip fabrication capacity, the cloud platforms — is overwhelmingly concentrated in the United States.

China is the only other country operating at anything approaching comparable scale. Its domestic AI champions — Baidu, Alibaba, Tencent, Huawei, and a growing roster of AI-native companies — are backed by state industrial policy that treats AI leadership as a national security priority of the first order. China’s approach is different from America’s, more state-directed and less reliant on private venture capital, but the ambition and the investment are real.

The European Union is trying, with genuine seriousness, to build a competitive position — but it is doing so primarily through regulation rather than investment, which is a strategy that shapes the environment without building the capability. Europe is writing the rules of a game that others are winning.

And then there is everyone else. The developing world — sub-Saharan Africa, most of South and Southeast Asia, Latin America, the Middle East outside the Gulf states — is largely watching this unfold from a position of almost complete dependency. These regions consume AI products built elsewhere, on infrastructure owned elsewhere, trained on data that overwhelmingly represents elsewhere. They are not building the technology. They are not shaping its development. They are not investing at the scale required to develop domestic capability. They are, in the precise historical sense of the term, falling behind.

Why It’s Happening: The Four Failure Modes

The countries missing the AI transition are not doing so randomly. There are four recurring failure modes that explain most of the gap, and each one has a direct parallel in the 19th century industrialization story.

The first is infrastructure deficit. Industrialization required coal, iron, and reliable energy. Countries without these resources or the infrastructure to deploy them were structurally disadvantaged from the start. AI requires reliable electricity, high-speed internet connectivity, and data center capacity. Roughly 2.6 billion people still lack reliable internet access. In sub-Saharan Africa, less than 40% of the population has regular internet connectivity. You cannot participate in an AI economy without the foundational infrastructure any more than you could participate in an industrial economy without reliable energy. The infrastructure gap is not a temporary inconvenience. It is a structural barrier that compounds every other disadvantage.

The second is the education and talent pipeline. Industrial nations built technical education systems — engineering schools, polytechnics, apprenticeship programs — that produced the human capital required to operate, maintain, and eventually innovate within industrial systems. Countries that lacked this pipeline couldn’t staff their own industrialization even when capital was available. The AI equivalent is computer science education, mathematics, and data literacy at scale. Most developing nations are producing tiny numbers of AI-capable graduates relative to their population size, and the best of those graduates emigrate to the United States, the UK, Canada, or the Gulf, creating a brain drain that simultaneously depletes the home country and enriches already-advantaged nations.

The third failure mode is elite capture and political resistance. In 19th century non-Western nations, the groups with the most political power — landed aristocrats, traditional merchants, religious institutions — often had the most to lose from industrialization. A textile magnate who had built his position on hand-weaving networks had no incentive to support the mechanization that would destroy his competitive advantage. These groups consistently used their political influence to slow, dilute, or redirect industrialization efforts. Today, the equivalent groups are traditional industries, incumbent businesses, and political elites whose power is rooted in existing economic structures. The resistance to AI adoption in many developing nations is not random. It reflects the interests of people who benefit from the status quo.

The fourth failure mode is the most insidious: the illusion of adequacy. Many countries in the 19th century believed they were keeping pace because they could see and use the products of industrialization — they bought British textiles, rode on British-built railways, used British-manufactured goods. They confused consumption of industrial products with participation in industrialization. They were customers of the new economy, not builders of it. The distinction only became clear when the terms of trade shifted permanently against them.

Today, a country whose population uses ChatGPT, whose businesses subscribe to Microsoft Copilot, and whose government is piloting AI chatbots for citizen services is consuming AI. It is not building AI capability. The distinction feels minor today. It will feel catastrophic in twenty years, for the same reason it felt catastrophic to the Ottoman Empire in 1900.

The Compounding Gap

What made the 19th century divergence so durable was not the initial gap. It was the compounding. Industrial nations used their productivity advantage to fund research, education, and infrastructure, which increased their productivity advantage further, which funded more research and infrastructure. The gap did not stay constant. It widened, decade by decade, because advantage begets advantage.

AI creates the same compounding dynamic, and it does so faster. The nations and companies training large AI models are generating data about how those models are used, which is fed back into training better models, which attract more users, which generate more data. The feedback loop operates at digital speed. The compounding is not generational, as it was with industrialization. It is annual.

There is a specific mechanism by which this compounding becomes particularly difficult to reverse: the talent concentration loop. The best AI researchers in the world want to work where the best infrastructure, the best datasets, and the most ambitious problems are. That is currently the United States and, to a lesser extent, China and the UK. As talent concentrates in these locations, it produces better AI, which attracts more talent, which produces better AI. Countries outside this loop do not just fall behind. They actively lose the people who could help them catch up.

India is the most striking illustration of this dynamic. It produces a remarkable number of world-class engineers and computer scientists — and exports the majority of them to Silicon Valley, Seattle, and New York. The Indian diaspora is a central pillar of American AI development. Meanwhile, India’s domestic AI ecosystem, while growing, remains dependent on foreign models, foreign cloud infrastructure, and foreign capital. India is, simultaneously, one of the world’s largest contributors to AI development and one of the world’s largest examples of brain drain preventing domestic AI capability from accumulating.

The Countries Getting It Right — And What They Know

Not every developing or mid-tier nation is making this mistake. A handful of countries have read the historical pattern clearly and are acting on it with urgency.

The United Arab Emirates has moved with a speed that is extraordinary for a nation of its size. It has established a national AI strategy, created the world’s first national minister of artificial intelligence, funded the development of Falcon — one of the most capable open-source large language models in the world — and is investing heavily in AI education and infrastructure. The UAE is a small country with enormous capital and a leadership that has studied what happened to oil-dependent economies that failed to diversify before the energy transition. It is determined not to repeat that pattern with AI.

Singapore has built an AI governance framework and investment strategy that positions it as the regional hub for AI development in Southeast Asia. Its government has understood since the 1980s that a small city-state with no natural resources survives by being maximally capable and maximally connected. AI is simply the latest expression of that strategic logic.

Rwanda, remarkably, has emerged as one of the more thoughtful AI policy actors in sub-Saharan Africa. It has established AI research initiatives, partnered with global institutions, and is attempting to use AI specifically to leapfrog infrastructure deficits in healthcare and agriculture. It is a small-scale example, but it demonstrates that the constraint is political will and strategic clarity, not geography or culture.

What these countries share is not simply money. They share a clear-eyed understanding of what general purpose technologies do to the distribution of global power — and a leadership class that has internalized the historical lesson that missing the transition is not a recoverable mistake on a human timescale.

What Needs to Happen — And Why It Probably Won’t

The path for developing nations to avoid the second divergence is not mysterious. It requires investment in digital infrastructure — electricity, connectivity, data centers — as a national priority equivalent to road and port construction in the industrial era. It requires education reform that treats computational thinking, mathematics, and AI literacy as foundational skills rather than specialist knowledge. It requires policies that retain talent rather than creating conditions that guarantee emigration. And it requires industrial policy that supports domestic AI development — not necessarily at the frontier, but at the level of application and adaptation to local problems and local languages.

None of this is technically difficult to understand. All of it is politically difficult to execute. The same forces that slowed industrialization in the 19th century — elite resistance, institutional inertia, the prioritization of short-term stability over long-term capability — are operating in the same direction today.

There is also a structural problem that did not exist in the industrial era: the technology is moving faster than institutional capacity to respond. The industrialization transition unfolded over roughly a century. The AI transition is unfolding over a decade. The window for catching up is narrower, the compounding is faster, and the gap between action and consequence is shorter. Countries that took thirty years to begin industrializing in the 19th century were damaged but recoverable. Countries that take thirty years to begin serious AI adoption in the 21st century may find the gap has become structural in a way that decades of subsequent effort cannot close.

The Question History Will Ask

In 1850, a thoughtful observer looking at the Ottoman Empire, the Qing Dynasty, or Mughal India could have described, with reasonable precision, what those civilizations needed to do to avoid the divergence that was coming. The analysis was not hidden. The trajectory was visible. The choices were available.

The choices were not made — not at the scale, the speed, or the consistency required. And the world that resulted from those unmade choices is the one we still live in. The inequality between nations that shapes every aspect of global politics today — the migration crises, the trade tensions, the debt dependencies, the military imbalances — traces its roots directly to decisions made, and not made, between 1800 and 1900.

We are in that window again. The decisions being made right now — about AI investment, about education, about infrastructure, about talent retention, about industrial policy — will shape the distribution of global wealth and power for the next hundred years with the same determinism that industrial decisions shaped the last hundred.

The countries that are acting with urgency are not panicking. They are reading history. The countries that are watching, waiting, or treating AI as a distant concern for richer nations are not being cautious. They are repeating the most consequential mistake of the modern era.

History does not offer many second chances to avoid a divergence. The question is not whether the gap will open. It is already opening. The question is which side of it each country will be on — and whether the people with the power to make that choice understand what they are actually deciding.

Written at the intersection of economic history, technology, and global power.