A worker in Milan earns, on average, roughly €38,000 a year. A worker doing a comparable job in Enna, a small city perched on a hilltop in central Sicily, earns closer to €17,000. That is not a comparison between two different countries, two different legal systems, or two different currencies. It is a comparison within a single nation, under a single government, separated by a train ride of about twelve hours. Both workers are Italian citizens. Both are entitled to the same constitutional protections, the same public healthcare, the same passport. Yet one of them lives in an economy that looks, by the numbers, like northern Europe, and the other lives in an economy that resembles parts of southern Portugal.
This kind of gap is not a curiosity. It is a pattern. It repeats itself across every continent, at every scale: between countries, between regions, between the neighbourhoods on opposite sides of a river. Wealth does not spread evenly across space, and it never has. The question is why. Why does economic prosperity seem to cluster in some places and abandon others? Why do some cities become magnets for talent and capital while towns fifty kilometres away empty out? And why, despite decades of policy effort, does the pattern prove so stubbornly difficult to change?
These are not rhetorical questions. Economics has detailed, often surprising answers to each of them. This article will walk through those answers: the forces that cause wealth to concentrate, the traps that prevent poor places from catching up, the role of history and institutions in shaping the map of prosperity, and what any of this means for a young person choosing where to build a career.
Figure 1: GDP Per Capita By Italian Region, 2022
The View from 30,000 Feet
Before zooming in on mechanisms, it is worth pausing on the sheer scale of global wealth inequality across space. The richest country in the world by GDP per capita (PPP), Luxembourg, produces roughly $140,000 of economic output per person per year. Burundi, one of the poorest, produces about $850. That is not a gap of two or three times. It is a factor of more than 160. A Luxembourgish worker produces in three days what a Burundian worker produces in an entire year.
These figures are not just abstract accounting. They translate directly into life expectancy, infant mortality, educational attainment, and the basic material texture of daily existence. A child born in Norway can expect to live past 80. A child born in Chad may not reach 55. The geography of wealth is, in other words, a geography of almost everything else that matters.
What drives these differences? The easy answers, the ones that circulate in casual conversation, tend to fall into two camps. One says that rich countries are rich because of natural resources or favourable climates. The other says it is simply a matter of culture or effort. Both contain their share of truth, but neither holds up well under scrutiny. Resource-rich countries like the Democratic Republic of the Congo remain desperately poor, while resource-scarce places like Singapore or Switzerland are among the wealthiest on earth. Culture matters, but it is shaped by history, and history is shaped by institutions, and institutions are shaped by power, and power is shaped by economics. The causal arrows run in circles. Untangling them is the project of an entire academic field.
What is clear, though, is that the geography of wealth operates at multiple scales simultaneously. There is a gap between rich continents and poor ones, but also between rich and poor countries on the same continent, between rich and poor regions within the same country, and even between rich and poor postcodes within the same city. The forces at work are not identical at every scale, but they rhyme.
Figure 2: GDP Per Capita (PPP, 2023): A Global Snapshot
The Gravity of Cities: Agglomeration Effects
The most powerful explanation for why wealth clusters in space comes from a concept that economists call agglomeration effects. The term is technical, but the idea behind it is not. In essence: economic activity tends to attract more economic activity. Places that are already productive tend to become more productive, because concentrating people, firms, and ideas in the same location generates advantages that would not exist if everyone were spread out evenly.
This is the engine that powers the world’s great cities. It is the reason London, New York, and Tokyo are not merely large but disproportionately wealthy relative to the countries that contain them. And it operates through three main channels that are worth understanding individually.
The first is labour market pooling. When many firms in related industries locate near one another, they create a deep, shared pool of workers with relevant skills. This benefits both sides. Workers can switch jobs without relocating, which makes them more willing to invest in specialised skills. Firms can hire and adjust their workforce more easily, which reduces risk. A software engineer in San Francisco can lose her job on a Monday and find a new one by Friday, because the pool of potential employers is enormous. A software engineer in a small city with only one or two tech firms faces a far more precarious situation.
The second channel is knowledge spillovers. Ideas, it turns out, do not travel as easily as we often assume. Formal knowledge, the kind written down in papers and patents, can move anywhere. But tacit knowledge, the practical, intuitive understanding of how to do things well, tends to spread through face-to-face interaction, informal conversation, and proximity to people who are working on similar problems. This is why industries cluster: filmmaking in Los Angeles, finance in London, fashion in Milan. The people in these clusters learn from one another, often without realising it, simply by being embedded in the same professional ecosystem. The economist Alfred Marshall, writing in 1890, described this phenomenon with an elegant phrase: the “mysteries of the trade” are, as it were, “in the air.”
The third is supplier and customer proximity. Firms that locate near their suppliers and their customers save on transport costs, reduce delivery times, and can coordinate more easily. This creates a self-reinforcing logic: a car manufacturer attracts component suppliers, the presence of component suppliers attracts another car manufacturer, and so on. The result is the kind of industrial cluster that defines regions like the Stuttgart area in Germany or the motor valley of Emilia-Romagna.
What makes agglomeration effects so powerful, and so important for understanding the geography of wealth, is that they are self-reinforcing. Once a city reaches a critical mass of economic activity, the advantages of being there increase, which attracts more firms and workers, which further increases the advantages. Economists call this a positive feedback loop, or, in plainer language, a snowball effect. It helps explain one of the most robust empirical findings in urban economics: doubling a city’s population is associated with a roughly 10 to 15 percent increase in productivity per worker, a relationship that holds across countries and time periods.
This means that big cities are not just bigger versions of small cities. They are qualitatively different economic environments. A worker who moves from a town of 100,000 to a city of a million is not simply joining a larger labour market; she is entering a setting where every worker, on average, produces more. The city itself is a productivity multiplier.
Figure 3: City Size and Productivity: The Urban Scaling Relationship
The Sharpest Edge: San Francisco and the American Divide
Few places on earth illustrate the power of agglomeration more starkly than the San Francisco Bay Area. The five-county region centred on San Francisco and Silicon Valley, home to roughly 4.7 million people, produced a GDP of approximately $764 billion in 2022, making it one of the twenty largest economies in the world measured as a standalone entity. Average annual wages in San Jose-Sunnyvale-Santa Clara, the metropolitan area that encompasses Silicon Valley, exceeded $145,000 in 2023, more than double the national average of roughly $68,000. In San Francisco proper, the figure was comparable.
Now consider Lumberton, a small city of about 20,000 people in Robeson County, North Carolina. Robeson County's median household income sits around $33,000 a year, roughly a quarter of what a single worker earns, on average, in Silicon Valley. Life expectancy in Robeson County is about 72 years; in San Mateo County, in the heart of the Bay Area, it is closer to 84. The two places are separated by less than 4,000 kilometres and zero legal barriers to movement. The educational divide compounds the economic one: bachelor's degree attainment in Robeson County stands at 13.8%, compared to 55.2% in Silicon Valley, a gap shaped in part by access to universities, scholarship systems, and the intergenerational transmission of educational capital."
The Bay Area's extraordinary wealth is not an accident of natural resources or geography. It is agglomeration in its most concentrated form. What began with Stanford University's post-war technology transfer programmes in the 1950s became, over successive waves of innovation, a self-reinforcing cluster of venture capital, engineering talent, and institutional knowledge that no other region has been able to replicate at scale. The labour market pooling is enormous: a machine learning engineer who leaves Google can walk into a role at Meta, Apple, or any of several thousand startups without changing her commute. The knowledge spillovers are intense: ideas circulate through networks of alumni, former colleagues, and casual connections in a way that simply cannot be reproduced by Zoom calls. And the supplier ecosystem, from specialised law firms to chip fabricators, is unmatched.
Robeson County, meanwhile, is a near-textbook case of the opposite trajectory. Its economy was historically anchored in tobacco farming, textile manufacturing, and military employment from a nearby base. As each of these contracted, no replacement emerged with sufficient mass to trigger a new agglomeration cycle. The skilled left. Services deteriorated. The feedback loop ran in reverse.
What makes this comparison so instructive is that it is not a story about individual effort or talent. There are ambitious, intelligent people in Robeson County, just as there are in Palo Alto. The difference is the economic environment surrounding them: the density of opportunity, the depth of networks, the thickness of the labour market. The geography of wealth is not about who people are. It is about where they are, and what forces that location sets in motion.
Figure 4: Two Americas: Silicon Valley vs. Robeson County, NC
Why Struggling Places Stay Stuck
If agglomeration effects explain why prosperous places tend to become more prosperous, a natural question follows: what happens at the other end? If concentration is self-reinforcing on the way up, is decline equally self-reinforcing on the way down?
The answer, unfortunately, is yes. The same feedback logic that creates booming cities also creates what economists call spatial poverty traps. When a region loses its primary source of employment, whether through deindustrialisation, technological change, or a shift in global trade patterns, the consequences cascade. Workers leave to find jobs elsewhere. Their departure shrinks the local tax base, which reduces the quality of public services: schools, transport, healthcare. The deterioration of services makes the place less attractive, which causes more people to leave. Shops close. Property values fall. The skilled and the young go first, leaving behind an older, less mobile population. The cycle feeds on itself, and without a significant external intervention, it rarely reverses spontaneously.
The textbook case is the deindustrialisation of northern England. Towns like Sunderland, Hartlepool, and Burnley were built around coal, steel, and shipbuilding. When those industries contracted in the 1970s and 1980s, accelerated by policy choices under the Thatcher government, the towns did not simply transition to new industries. They hollowed out. Between 1979 and 1990, the North East of England lost roughly a third of its manufacturing employment. Decades later, GDP per capita in the North East remains among the lowest of any English region, and the gap with London has, if anything, widened rather than narrowed.
Italy has its own version of this story, and it is far older. The Mezzogiorno, the southern third of the country, has been persistently poorer than the north for well over a century. GDP per capita in Calabria is less than half that of Lombardy. Southern unemployment rates are roughly double the national average. The Italian government has spent billions of euros over decades trying to close this gap, through infrastructure investment, tax incentives, and direct transfers, with limited success. The gap has narrowed at times and widened at others, but it has never closed, a fact that should give pause to anyone who believes that regional inequality is a problem that simply requires more spending.
Figure 5: A Gap That Will Not Close: GDP Per Capita in Northern vs. Southern Italy, 1950-2022
Figure 6: The Anatomy of Regional Decline
Ghosts in the Machine: History and Institutions
The forces described so far, agglomeration on one side, self-reinforcing decline on the other, are powerful. But they do not operate in a vacuum. They interact with something deeper: the institutional and historical inheritance of a place. And this inheritance can be astonishingly persistent.
Consider again the Italian North-South divide. Economists and political scientists have long debated its origins, and one of the most influential accounts comes from Robert Putnam’s study of Italian regional governments in the 1990s. Putnam found that the quality of regional governance in Italy, measured by efficiency, responsiveness, and the provision of public goods, varied enormously, and that this variation correlated strongly with differences in civic culture: the density of voluntary associations, newspaper readership, voter turnout, and social trust. Northern regions scored high on these measures; southern regions scored low.
What made Putnam’s finding remarkable was its historical depth. The civic traditions of northern Italy, he argued, could be traced back to the free city-states and communal republics of the medieval period: Florence, Bologna, Venice. These places had centuries of experience with self-governance, horizontal cooperation, and civic participation. Southern Italy, by contrast, had spent much of its history under centralised, authoritarian rule, first by the Normans, then by successive foreign monarchies, which left behind a legacy of vertical dependence, low social trust, and weak civic institutions. The implication was striking: institutional differences established 700 years ago were still shaping economic performance at the end of the twentieth century.
At the global scale, a similar argument has been made about the legacy of colonialism. Economists Daron Acemoglu, Simon Johnson, and James Robinson, in a landmark 2001 paper, argued that the type of colonial institutions established by European powers had lasting effects on economic development. In places where colonisers found low mortality rates and settled permanently (like Australia and Canada), they built inclusive institutions: property rights, rule of law, constraints on executive power. In places where mortality was high and permanent settlement unattractive (like much of sub-Saharan Africa and tropical Latin America), they built extractive institutions: designed to funnel resources out of the colony with minimal investment in local governance or human capital. Centuries later, the institutional differences persisted, and with them, the economic divergence.
This is an economic argument about path dependence: the idea that initial conditions can lock places into trajectories that are extremely difficult to escape, because institutions shape incentives, incentives shape investment, and investment shapes growth, in a cycle that compounds over generations.
Figure 7: Colonial Origins and Modern Wealth: Settler Mortality vs. GDP Per Capita
The Future of the Map
Everything this article has described so far, the clustering of wealth, the feedback loops that entrench it, the stubborn persistence and roots of the gap between rich places and poor ones, happened in an era when the world’s most valuable industries still required physical infrastructure on a massive scale: factories, shipping routes, power grids. One of the defining industries of the next several decades, artificial intelligence, requires almost none of that. It requires talent, compute, capital, and proximity to other people working on the same problems. In other words, it requires exactly the conditions that agglomeration theory predicts will concentrate in a vanishingly small number of places. And that is precisely what is happening.
In 2023, the San Francisco Bay Area absorbed roughly half of all global venture capital invested in artificial intelligence, a figure of approximately $74 billion. Not half of American AI investment. Half of the world’s. The leading frontier AI laboratories, OpenAI, Anthropic, Google DeepMind, Meta’s FAIR division, are overwhelmingly headquartered in or around San Francisco. The researchers who build the most capable models circulate between these organisations in a labour market so dense and so specialised that it functions like a small town: everyone knows everyone, and a breakthrough at one lab diffuses through the ecosystem within weeks.
This is agglomeration at its most extreme, and it is intensifying rather than dispersing. The reason is structural. Training a frontier AI model is not like building a mobile application. It requires unprecedented sums in compute, access to proprietary datasets, and teams of researchers whose expertise exists in perhaps a few thousand people globally. Those people want to work and live near one another, because the knowledge spillovers in AI research are enormous and overwhelmingly tacit: the kind of intuition about model architecture, training dynamics, and scaling behaviour that cannot be written down in a paper but spreads effortlessly through hallway conversations and job switches. The venture capitalists who fund these labs want to be close enough to take meetings in person. The cloud computing providers who supply the hardware are building data centres within reach. Every element of the ecosystem reinforces the others. The snowball is rolling faster, not slower.
London and Beijing represent secondary nodes, each with genuine strengths: London in AI safety research and DeepMind’s operations, Beijing in applied AI and state-backed compute infrastructure. A handful of other cities, Toronto, Paris, Tel Aviv, have credible AI clusters. But credibility is not competitive at the frontier. The gap between San Francisco and everywhere else in AI is not narrowing. By most measures, it is widening. The question is no longer whether AI development will concentrate. It already has. The question is what that concentration means for everywhere else.
Figure 8: Where the AI Money Goes: Global AI Venture Capital by City, 2023
AI is the sharpest case, but it is part of a broader pattern. Across the knowledge economy, from finance to biotech to advanced manufacturing R&D, the highest-value activity is sorting itself into a shrinking number of city-regions. The economist Enrico Moretti has documented this divergence in granular detail: every high-tech job created in a city generates roughly five additional local service jobs, from baristas to doctors, meaning that the places that win the innovation economy win everything else as well. Richard Florida, who coined the term “superstar cities,” argues that we are heading toward a global urban hierarchy in which perhaps fifteen to twenty city-regions capture a disproportionate and growing share of talent, capital, and economic output, while everywhere else, including cities that are prosperous by historical standards, settles into a secondary role.
The evidence supports this. OECD data shows that the productivity gap between the most productive and least productive regions within developed countries has widened, not narrowed, over the past two decades. London’s economy has pulled further from Birmingham’s. Milan’s has pulled further from Naples’s. San Francisco’s has pulled further from everywhere. The places that were already winning the agglomeration game in 2000 have, for the most part, won it more decisively by 2025. The convergence that economists once expected, the idea that poorer regions would gradually catch up to richer ones as capital and technology diffused, has not materialised. In many countries, the opposite has occurred.
What makes this trend genuinely new, and not simply a continuation of the urbanisation patterns of the last two centuries, is its interaction with the economics of intangible capital. The industries driving the superstar city phenomenon do not produce physical goods that could, in principle, be manufactured anywhere. They produce software, algorithms, financial instruments, research, and intellectual property: outputs that are weightless, infinitely replicable, and subject to winner-take-most dynamics. A factory in Stuttgart and a factory in Bratislava can both produce cars. But the returns to being the single best AI laboratory, or the dominant global financial centre, are of an entirely different magnitude. These are industries where being in the top cluster is not incrementally better than being in the second-tier cluster. It is categorically different. And that categorical advantage reinforces the geographic concentration, because the people and firms competing for those returns have every incentive to locate where the top cluster already exists.
The political implications are significant but beyond the scope of this article. What belongs here is the economic observation: the forces described in every preceding section of this piece, agglomeration, feedback loops, institutional path dependence, are not relics of industrial history. They are accelerating. The geography of wealth is not flattening. It is, if anything, developing sharper peaks and deeper valleys. The world that AI and the knowledge economy are building is one in which your coordinates matter more, not less, than they did a generation ago.
Figure 9: Diverging, Not Converging: Productivity Gap Between Top and Bottom Regions
All of which returns us, with a different weight, to the train ride between Milan and Enna.
The Map Is Not Flat
Return, for a moment, to the train ride between Milan and Enna. The twelve hours of track between those two cities pass through some of the most visually dramatic scenery in Europe: the flat, industrial Po Valley giving way to the hills of Tuscany, the Tyrrhenian coast, the straits of Messina, and finally the arid interior of Sicily. The landscape changes profoundly. So, as this article has tried to show, does the economy beneath it.
The gap between Milan and Enna is not an accident and not a mystery. It is the product of agglomeration forces that reward density and punish thinness; of feedback loops that amplify early advantages and deepen early disadvantages; of institutional legacies that stretch back, in Italy’s case, to the Middle Ages; and of policy choices that have sometimes helped and often failed. Understanding these mechanisms does not make the gap any less troubling. But it does make it legible. It turns a frustrating fact of life into a problem with identifiable causes and, potentially, addressable components.
The deeper question, and the one this article deliberately leaves open, is whether the future will look like the past. Will the geography of wealth continue to concentrate, with a handful of superstar cities pulling further away from everywhere else? Or will new technologies, new patterns of work, and new political pressures redistribute prosperity more evenly across space? The forces of agglomeration are powerful and historically durable. But they are not laws of physics. They are the product of human choices, human institutions, and human technologies, all of which can change.
The map of wealth is not fixed. But neither is it easily redrawn. That tension, between the gravity of place and the possibility of change, is one of the most important puzzles in economics. And it is playing out, right now, in every city, every region, and every country on earth.