Superstar Cities — Why Five Metros Took 90% of Tech Growth

The term superstar city entered economics in 2006, when Joseph Gyourko, Christopher Mayer and Todd Sinai used it to describe a small set of metropolitan areas where house prices had pulled away from the rest of the United States for half a century [1]. Between 1940 and 2000, real average house prices grew more than elevenfold in San Jose, ninefold in San Francisco and more than sixfold in Seattle, while outside the top ten metros they at most tripled [1]. ✓ Established Their explanation was simple and unsentimental: a limited supply of land in a few uniquely desirable places, combined with a growing national population of high-income households, produces a bidding contest that only the rich can win [1]. The superstars were defined not by their output but by their scarcity. That definition matters, because it locates the problem in the housing market rather than in the labour market, and it predicts that the winners will grow richer without necessarily growing larger.

The labour-market evidence has since caught up with the housing-market theory. Robert Atkinson, Mark Muro and Jacob Whiton found that between 2005 and 2017 the five leading innovation metros raised their combined share of national innovation employment from 17.6 percent to 22.8 percent [8]. ✓ Established The innovation sector in their definition comprises 13 of the most technology-intensive and research-intensive industries in the American economy — the industries that pay the highest wages and generate the most patents. That the gains came from so few places, while 343 metros saw such jobs decline, led the authors to propose a federal programme of growth centres worth up to $700 million a year in direct research funding for a select group of interior metros [8]. A version of that proposal was adopted three years later, in a form examined in Section 8.

Artificial intelligence has intensified the concentration rather than diffusing it. A 2025 Brookings mapping of the American AI economy classified only two metros — San Francisco and San Jose — as AI superstars, with unmatched strength across talent, innovation and adoption; the Bay Area alone accounted for 13 percent of all AI-related job postings, and the top 30 metros for 67 percent [9]. ✓ Established Capital follows the same map. According to PitchBook-NVCA data, US startups raised an estimated $339.4 billion in 2025, and roughly 62 percent of that went to California-based companies, up from 54.2 percent in 2024 and 46.9 percent in 2023 [29]. About 66 percent of 2025 US deal value went to AI and machine-learning companies, and those rounds cluster heavily in the Bay Area [29]. A technology that was supposed to make location irrelevant is, at least in its build-out phase, the most geographically concentrated industry of the decade.

Outside the United States, the concentration is often starker because it is national rather than regional. The OECD's Regions and Cities at a Glance 2024 reports that in 2022 labour productivity in a country's most productive regions was double that of its least productive, with gaps of four times or more in Colombia, Mexico and Romania [18]. ✓ Established Capital-city regions were the most productive regions in 20 of 33 member and accession countries and second in five others [18]. Eurostat's 2024 regional accounts show the same hierarchy inside the European Union: capital-city regions have the highest GDP per capita in 19 of the 22 member states that are divided into regions, with Eastern and Midland Ireland at 268.3 percent of the EU average and Luxembourg at 244.6 percent [24]. Banks, media groups and corporate headquarters cluster in capitals, and each worker there produces more measured value on average [24].

Yet the same OECD data contain a qualification that the superstar narrative often omits. Over the two decades to 2022, productivity levels across OECD regions converged slowly, with low-productivity regions growing at 1.3 percent a year against 1.0 percent for high-productivity ones [18]. ◈ Strong Evidence The divergence that dominates political debate is therefore not universal. It is concentrated in specific sectors — advanced software, artificial intelligence, venture finance — and in specific countries, especially the United States, the United Kingdom, France, Japan and South Korea, where one metropolitan region dominates the national economy. Even within those countries the picture is more mixed than headlines suggest: the UK's Office for National Statistics found that in 2023 London's productivity lead over other regions, while still large, was the narrowest since its records began in 1998 [19]. The concentration is real; its permanence is not a law of nature.

What unites these cases is a combination of three forces that this report examines in turn. The first is agglomeration: the measurable productivity gains that come from putting skilled people, firms and capital close together [30]. The second is housing scarcity: the regulatory and physical limits on how many people can share in those gains [1]. The third is sorting: the tendency of the highly educated to cluster in the same places, raising wages and amenities there while the less educated are priced out [12]. ◈ Strong Evidence Each force has been documented in peer-reviewed research. The disagreement among economists is not about whether these forces exist, but about how large the resulting losses are, who bears them and what public policy can do about them. Remote work, which many expected to dissolve the superstar premium, turns out to have done something narrower and stranger.

The agglomeration premium is one of the most robust findings in economics. Early empirical work by Sveikauskas, Segal, Moomaw and Tabuchi, summarised by Antonio Ciccone in the New Palgrave Dictionary of Economics, estimated that doubling city size increases productivity by between three and eight percent [30]. ✓ Established The mechanisms are usually grouped under three headings. Sharing means that firms in dense places can split the cost of specialised suppliers, infrastructure and services. Matching means that larger labour markets pair workers with jobs that fit their skills more precisely, and reduce the risk of moving for both sides. Learning means that ideas travel faster when people encounter each other at work, in cafés and at conferences. Edward Glaeser's 2011 book Triumph of the City made the popular case that these mechanisms make the city humanity's greatest invention, leaving its residents richer, smarter, greener, healthier and happier [33].

Innovation industries are where these mechanisms are strongest. Tacit knowledge — the kind that cannot be written down — is transmitted face to face, and venture capital remains a local business in which investors prefer to fund companies they can visit. The Brookings growth-centres research found that the innovation sector's concentration increased steadily after 2005, not because other regions stopped trying, but because the leading hubs attracted a disproportionate share of new firms, new graduates and new capital [8]. ✓ Established The 2025 AI mapping repeated the finding for the newest technology: beyond the two superstars, 28 star hubs formed a second echelon of uniformly strong ecosystems, followed by 14 emerging centres and 29 focused movers that excel in one area [9]. Concentration at the top coexists with a long, thin spread of capacity across the rest of the country.

Enrico Moretti's contribution was to show that the gains do not stay with the engineers. In The New Geography of Jobs, published in 2012, he used Census data covering eight million workers in 320 metropolitan areas over three decades to estimate that each new job in software, technology or the life sciences generates about five additional jobs in the local economy [10]. ◈ Strong Evidence Two of those five are professional positions, such as doctors and lawyers, and three are non-professional, such as waiters and shop assistants [10]. The multiplier for traditional manufacturing is much smaller. This is the economic logic that makes a superstar metro self-reinforcing: every innovation job pulls in service jobs, which raise local demand, which attracts more firms, which recruit more innovators. It is also the logic that makes losing such a cluster so expensive for the places left behind.

The question is how much of the measured urban premium is really the city, and how much is the people who choose to live there. David Card, Jesse Rothstein and Moises Yi addressed this with linked employer-employee data from the US Census Bureau's Longitudinal Employer-Household Dynamics programme, following workers who move between commuting zones [7]. ✓ Established They found that about half of the variation in mean wages across commuting zones reflects differences in worker ability, and the other half genuine place effects [7]. Local industry clusters are associated with higher wages, but differences in industry mix explain only a small fraction of the place effects. And the college wage gap is larger in bigger, higher-wage places mostly because more skilled graduates sort there: about 66 percent of the variation in that gap reflects differences in the relative skills of the two groups [7].

Rebecca Diamond's work added a dimension that wage data alone miss. Between 1980 and 2000, the rise in the US college wage gap coincided with college graduates concentrating in high-wage, high-rent cities [12]. ✓ Established Local labour demand drove the initial sorting, but it was then amplified by improvements in amenities that followed the graduates into those cities [12]. Diamond's structural model concluded that changes in wages, rents and these endogenous amenities increased welfare inequality between high-school and college graduates by more than the rise in the college wage gap alone suggests [12]. ◈ Strong Evidence In other words, the superstar city is not only a better place to earn; for those who can afford it, it is a better place to live, and that advantage compounds the earnings gap. Restaurants, schools, parks and safety follow purchasing power, and purchasing power in these metros is increasingly sorted by education.

Taken together, this literature establishes the economic case for superstar cities more firmly than their critics sometimes allow. Agglomeration economies are real, measurable and largest in the industries that drive long-run growth [30]. Concentration of innovation is not a market failure in itself; it is partly the market working as theory predicts [10]. The problem that the rest of this report addresses lies elsewhere. If proximity pays, the cost of access to proximity determines who gets paid. In the superstar metros, that cost is set in the housing market, and the housing market in those metros has been governed for half a century by rules that limit how many people can share in the premium [1]. The next section follows the consequences of those rules for the workers who once used migration as their route to higher pay.

For roughly 100 years, per capita incomes in poorer US states grew faster than those in richer states, and the gap between them narrowed steadily [4]. ✓ Established Economists attributed much of this convergence to migration: workers moved from low-wage places to high-wage places, which held wage growth down where labour became abundant and pushed it up where labour became scarce. Over the three decades to 2010, Ganong and Shoag found, that process slowed sharply [4]. The timing coincided with a dramatic decline in population flows to wealthy states. Their explanation, using a novel measure of land-use regulation drawn from state appellate court records, was that increasingly tight land-use rules in wealthy areas raised housing costs so much that migration stopped paying for low-skill, low-wage workers, while it still paid for the highly skilled [4].

The asymmetry is the crucial point. A lawyer who moves from a low-cost region to New York or San Francisco earns enough extra to absorb the higher rent and still come out ahead. A cleaner making the same move faces a wage gain that is far smaller in proportion and a housing bill that is much the same as the lawyer's. For the less skilled, the premium is consumed by rent [4]. Ganong and Shoag estimated that had convergence continued at its 1940-1980 rate, the increase in hourly wage inequality between 1980 and 2010 would have been roughly 10 percent smaller [4]. ◈ Strong Evidence They also found that in higher-income places where land-use regulations were not tightened, convergence continued at its historical rate [4]. The machine had not broken everywhere. It had been switched off in the specific places where opportunity was greatest.

The household arithmetic in today's superstar metros makes the mechanism concrete. A CNBC analysis of local incomes and home prices found that typical homebuyers in Los Angeles, San Francisco and San Diego would need to spend up to 78 percent of their income on housing to buy a median-priced home [32]. ✓ Established The conventional affordability threshold is 30 percent. Card, Rothstein and Yi's linked-data study reached a striking conclusion about the net value of moving: local housing costs at least fully offset local pay premiums, so workers who move to larger commuting zones gain no higher consumption net of housing [7]. ◈ Strong Evidence The premium is real in nominal terms; after rent, for the average mover, it disappears. That finding explains why migration toward opportunity has fallen even as the opportunity itself has grown, and why the people who still make the move are disproportionately those whose skills command the very highest wages.

Housing scarcity in these metros is in large part a policy choice. The original Gyourko-Mayer-Sinai paper treated land scarcity as partly geographic — oceans, bays and mountains constrain San Francisco in ways that do not apply to Dallas — but the regulatory component has grown with every decade [1]. Zoning that reserves most residential land for detached single-family homes, minimum lot sizes, height limits, parking requirements and discretionary review processes all reduce the number of homes that can be built in response to rising demand. Hsieh and Moretti, in the study discussed in Section 7, argued that the most productive cities, New York, San Jose and San Francisco, had adopted the most stringent restrictions, effectively limiting the number of workers who could access their high productivity [2]. ⚖ Contested Whatever the precise macroeconomic cost, the distributional effect is not in dispute: those already inside gain asset wealth while those outside are excluded.

The lock-out produces a peculiar geography. Homeowners in superstar metros have experienced some of the largest real-asset gains in modern economic history, much of it unearned in any productive sense; the value comes from scarcity created by rules that existing residents vote to maintain [1]. Renters in the same metros face rising costs without the offsetting asset gain. Potential migrants in lagging regions never arrive. The pandemic year exposed how much of the superstar population is footloose when given the choice: the counties of Manhattan and San Francisco each lost close to 10 percent of residents aged 25 to 54 between 2020 and 2021, according to analysis by Paige Ouimet of the Kenan Institute [16]. ✓ Established Many of those who left were not low-wage workers. They were high earners in information technology, finance and professional services whose jobs had suddenly become portable [16].

The lock-out also shapes what kind of growth a superstar metro produces. When housing cannot expand, the gains from new innovation jobs are capitalised into land values rather than into employment. The Moretti multiplier assumes that service workers can live near the jobs that the multiplier creates [10]; when they cannot, the multiplier weakens or is exported to exurban commuting belts hours away. Higher home prices crowd out lower-income households and sort the population by income, exactly as Gyourko, Mayer and Sinai predicted in 2006 [1]. ✓ Established The result is a metro that is richer per head, smaller than it would otherwise be, and increasingly dependent on long commutes, international immigration and public subsidy to staff its hospitals, schools and restaurants. The next section turns to the workers on the far side of that locked door.

David Autor's 2019 Richard T. Ely lecture to the American Economic Association, published as Work of the Past, Work of the Future, documented how the urban labour market for non-college workers changed over 45 years [5]. In 1970, non-college workers in dense commuting zones were about 25 percentage points more likely to work in middle-skill occupations, such as clerical, administrative, sales and production jobs, than their peers in low-density areas [6]. ✓ Established Those jobs paid well and offered a ladder. By 2015 the middle-skill employment share among non-college workers was several points lower in dense commuting zones than in low-density ones [6]. Automation and international trade had eliminated precisely the mid-skilled jobs that had been disproportionately located in cities, and the non-college urban workforce had been pushed down into the same low-wage service occupations available everywhere else [5].

The consequence for wages was direct. Now that non-college urban workers hold the same jobs as their rural and suburban peers, in custodial work, food service, security, personal care, transport and labouring, the urban wage premium for non-college workers has all but vanished, in the words of the NBER Digest summary of Autor's findings [6]. ✓ Established The urban premium for college graduates, by contrast, has held up [5]. This is the hollowing out of the middle-skill labour market in its geographic form. A city that once offered upward mobility to workers without degrees now offers them higher rents, similar wages and fewer routes upward. The superstar metro is the place where the returns to education are steepest, and therefore the place where the penalty for lacking a degree is most acutely felt in the cost of living [7].

Mobility data add a complication that pure wage data miss. Raj Chetty, Nathaniel Hendren, Patrick Kline and Emmanuel Saez, using tax records on more than 40 million children and their parents, found enormous variation in upward mobility across American commuting zones [11]. ✓ Established The probability that a child born into the bottom fifth of the income distribution reaches the top fifth was 12.9 percent in San Jose but only 4.4 percent in Charlotte [11]. High-mobility areas shared five features: less residential segregation, less income inequality, better primary schools, greater social capital and more family stability [11]. The superstar metros, in other words, can be engines of mobility for the children who grow up inside them. The lock-out matters precisely because it prevents more children from growing up there, and because the families who leave in search of lower rents take their children out of the places where mobility is highest.

The political consequences of this geography have become impossible to ignore. Andrés Rodríguez-Pose of the London School of Economics argued in 2018 that the so-called places that don't matter, from declining industrial towns to rural areas and lagging regions, had begun to revolt against the status quo through the ballot box, in a wave of populism with territorial rather than purely social foundations [26]. ◈ Strong Evidence The pattern he described ran across the developed world, with anti-system support concentrated in places that had lost jobs, young people and status [26]. Rodríguez-Pose's prescription was not to abandon agglomeration but to pursue place-sensitive development that taps unused potential in every region, rather than relying only on transfers to compensate those left behind [26]. It is a direct challenge to the view that the efficient response to regional decline is simply to encourage people to move.

The demographic drain compounds the economic one. When young, educated workers leave lagging regions for superstar metros, they take with them the tax base, the entrepreneurial energy and the future families those regions need. Japan offers the clearest example: in 2024 Tokyo recorded a net inflow of 79,285 people from elsewhere in Japan, and the largest group by far were young people aged 20 to 24, with a net inflow of 86,908 [21]. ✓ Established At the other end of working life, people aged 55 to 64 left Tokyo on net [21]. The capital draws in people at the start of their careers and releases them near the end, while the regions bear the cost of raising and educating them and later of caring for them. The same pattern, at a smaller scale, is visible in the movement of graduates toward London and toward Seoul [19][22].

None of this means that superstar cities are a net harm. Their productivity funds national tax systems that redistribute substantial resources to lagging regions, and their universities and firms produce innovations that diffuse nationally over time [10]. But the distributional structure of the gains has shifted. The urban premium was once a broadly shared prize, open to anyone willing to move. It is now a prize largely reserved for those who already hold a degree, an inherited home or both. Diamond's finding that sorting by skill amplified welfare inequality beyond the wage gap alone [12] ◈ Strong Evidence and Autor's finding that the non-college urban premium has vanished [5] describe the same transformation from two directions. The city as escalator has become, for many, the city as gated community.

The United Kingdom has one of the most pronounced regional productivity gaps in the developed world. In 2023, output per hour worked in London was 28.5 percent above the UK average and the South East was 7.7 percent above; every other region was below average, with Wales, the West Midlands, the East Midlands and the North East between 14.6 and 15.1 percent below [19]. ✓ Established The Office for National Statistics noted, however, that the gap between London and the other regions was at its lowest level since 1998 [19]. The 2022 Levelling Up White Paper set 12 missions to be achieved by 2030, from pay and productivity to transport, broadband and pride in place. The Institute for Government judged that only four of the 12 were clear, ambitious and properly measured, and concluded that as drafted they would not reduce regional inequality [27]. ◈ Strong Evidence

France concentrates economic power in its capital region to a degree unusual among large European economies. INSEE reported that Île-de-France produced 31 percent of the wealth created in metropolitan France, well above its 19 percent share of the population, and that this share had risen by several percentage points in preceding years [20]. ✓ Established INSEE attributed the region's productivity lead primarily to the concentration of highly qualified employment: managers, engineers and specialists in finance, technology and corporate headquarters functions [20]. Paris therefore displays the classic superstar profile, with high productivity driven by skilled workers, very high housing costs and a political geography marked by resentment in peripheral and rural areas of the kind Rodríguez-Pose describes [26]. The French state has long attempted to build regional counterweights, yet the capital's share of output has continued to edge upward rather than fall [20].

Japan is the most sustained national experiment in trying to reverse capital concentration. For more than a decade, the dominance of Tokyo has been the target of regional revitalisation policy, yet in 2024 the capital's net inflow of residents from elsewhere in Japan reached 79,285, about 11,000 more than the year before [21]. ✓ Established Only six other prefectures recorded net inflows, three of them, Kanagawa, Saitama and Chiba, forming part of the greater Tokyo commuter belt [21]. In 2023 the government raised its relocation grant for families leaving the Tokyo area to one million yen per child, conditional on living in the new region for at least five years and either finding a job or starting a business there [31]. ✓ Established The grant is small relative to the economic pull of the capital, and the migration data suggest it has not altered the trend [21].

South Korea presents the most extreme case of capital concentration among large OECD economies. As of December 2023, Seoul, Incheon and Gyeonggi Province together had 26.1 million registered residents, 50.7 percent of the national population, and 51.6 percent of the working population [22]. ✓ Established The capital area also held 49.1 percent of all businesses in 2022 [22]. Korea's response has been the most radical of any democracy: the construction of Sejong City, about 120 kilometres south of Seoul, as an administrative capital. Since 2012, 44 central government bodies, including the finance and health ministries, have moved there, and in September 2026 the government announced plans to relocate about 350 more ministries and agencies from the Seoul region starting in 2027, alongside a second presidential office in Sejong by 2029 [23]. ✓ Established The population of Seoul proper is falling, but Gyeonggi and Incheon continue to grow [22].

Germany is the instructive counter-example. Its economy is polycentric: Munich, Frankfurt, Hamburg, Stuttgart, Berlin, Cologne and Düsseldorf all host major corporate headquarters, and no single metro dominates national output. Eurostat's regional accounts put Hamburg at 196.1 percent of EU-average GDP per capita in 2024, but Hamburg is a port city-state rather than the national capital [24]. ✓ Established Research by the ifo Institute found that regional inequality in Germany and the EU, measured by the dispersion of GDP per inhabitant across districts, has been decreasing since 2000, and that it is considerably lower when measured by output per employee, because commuting allows residents of many districts to work in nearby production hubs [25]. ◈ Strong Evidence Federalism, distributed universities and a strong base of regional manufacturers all contribute. The German case shows that concentration is not inevitable, though its structure was set by history rather than by recent policy.

The five cases share a pattern with the United States. Productivity and capital concentrate where skilled workers already are; housing scarcity in those places prices out the less skilled; and the political response arrives late and is modest relative to the forces involved [18]. The key difference is the scale of the dominant metro relative to the nation. In the United States, superstar status is shared by several metros on two coasts, and the country is large enough that Austin, Raleigh and Nashville can emerge as challengers [16]. In Korea, France, Japan and the UK, the dominant metro is the capital, and policy toward it is entangled with national identity, the location of government and the distribution of political power. The timeline below traces how the research and the policy responses developed over eight decades.

The pandemic created the largest natural experiment in economic geography in a century. Almost overnight, a large share of high-wage knowledge work became independent of location, and the people doing it had both the means and the motive to leave expensive metros. Many did. Paige Ouimet's analysis for the Kenan Institute found that from 2020 to 2021 the counties of Manhattan and San Francisco each lost close to 10 percent of residents aged 25 to 54, while Austin, Jacksonville, Raleigh and Charlotte recorded large inflows [16]. ✓ Established High-skill workers in information technology, finance and professional services drove the redistribution [16]. For a moment it appeared that the superstar premium might dissolve, as skilled workers carried superstar salaries to cheaper places, spent them locally and took the Moretti multiplier with them [10].

Five years later, the persistent effect looks far more local. Arjun Ramani, Joel Alcedo and Nicholas Bloom, in research published in PNAS in December 2024, found that the 12 largest US cities had lost about 8 percent of their downtown residents since the pandemic, and that about 60 percent of the households that left moved to suburbs of the same metro [13]. ✓ Established Hybrid work, with two or three days a week in the office, permits a longer commute but not a move to another region [13]. Barrero, Bloom and Davis estimate that working from home has stabilised at about a quarter of paid workdays among Americans aged 20 to 64 [15]. ✓ Established That is a permanent shift, but it is mostly hybrid, and hybrid work ties employees to the metro where the office is. The donut effect hollowed out the centre; it did not move the metro.

The superstar metros also recovered faster than expected. The US Census Bureau's Vintage 2024 estimates, released in March 2025, showed that every metropolitan area with more than five million residents grew in the year to July 2024 [17]. ✓ Established The New York metro, at 19.9 million people, added more residents than any other: about 147,000 people moved out to other parts of the country, but nearly 288,000 immigrants moved in [17]. The domestic outflow from the largest metros continued, but international migration more than offset it, and immigrants settled disproportionately in urban cores while domestic movers headed for distant suburbs [17]. ◈ Strong Evidence The population of the superstar metros, in other words, is being replenished from abroad rather than from the rest of the country, a pattern that reinforces rather than resolves the lock-out of domestic lower-wage workers.

The most important effect of remote work may have been on house prices everywhere. John Mondragon and Johannes Wieland estimate that the shift to remote work explains more than half of the 18.9 percent rise in US real house prices between 2019 and 2023 [14]. ◈ Strong Evidence Using variation in remote-work exposure across metros, they find that each additional percentage point of remote work raised house prices by 0.92 percent, through higher demand for home space rather than through speculation [14]. Residential rents rose and commercial rents fell. The redistribution that remote work achieved was therefore not of opportunity but of housing pressure: it exported superstar-style price increases to suburbs, exurbs and amenity towns, making housing less affordable in many of the places that had been the escape valve for workers priced out of the superstars.

Remote work did not reverse the concentration of the innovation economy either. If anything, the AI boom that followed reinforced it: the 2025 data show the highest share of US venture capital going to California on record, at roughly 62 percent [29], ✓ Established and the Brookings AI mapping classifies only San Francisco and San Jose as superstars, with the Bay Area alone accounting for 13 percent of AI-related job postings [9]. The work that is easiest to do remotely, such as routine coding, analysis and support, is also the work most exposed to automation, while the work that commands the largest premium continues to depend on the tacit-knowledge networks that dense clusters provide [30]. Distance work spread the routine; proximity kept the scarce.

The remote-work test therefore produced a mixed verdict. It proved that a substantial share of high-wage work can be done anywhere, and some households did use that freedom to leave superstar metros for cheaper regions; Austin, Raleigh and Nashville gained [16]. But the dominant pattern was intra-metropolitan: the donut, not the diaspora [13]. ◈ Strong Evidence Remote work weakened the downtown, not the metro. It raised housing costs nationally rather than equalising them [14], and it did nothing for workers whose jobs require physical presence: the cleaners, carers, cooks and drivers whose exclusion from superstar metros is the core of the lock-out. A technology that frees the already advantaged from geography does not, by itself, give geography back to the disadvantaged.

In 2015 Chang-Tai Hsieh and Enrico Moretti circulated a paper that became the single most influential number in the housing-supply debate. Using a spatial equilibrium model covering 220 metropolitan areas, they estimated that housing constraints in high-productivity cities lowered aggregate US growth by 36 percent between 1964 and 2009 [2]. ⚖ Contested The final published version, in the American Economic Journal: Macroeconomics in 2019, framed the counterfactual as the gain from relaxing land-use regulation in New York, San Jose and San Francisco to the level of the median US city [2]. The figure was quoted by economic advisers, housing advocates and journalists across the political spectrum, and it underpinned the argument that zoning reform in a few metros was among the most powerful growth policies available to any government.

In April 2026 the same journal published a comment by the economist Brian Greaney. Attempting to replicate the paper, he found that the published counterfactual, which the authors summarised as raising GDP by 3.7 percent, would in fact lower output, and he documented errors in the original code that explained the discrepancy [3]. ✓ Established He further showed that the model's results depend on the arbitrary choice of population unit, and proposed a modification that removes that dependence. In the corrected model, relaxing land-use rules in the three cities still raises output, but by an amount two orders of magnitude smaller than reported [3]. ✓ Established The episode is an object lesson in how a single striking estimate can carry a policy debate for a decade before anyone attempts to reproduce it, and in how much weight a public argument can rest on code that few readers ever inspect.

The correction does not settle the debate; it relocates it. Greaney's result concerns one model's estimate of aggregate output. It says nothing directly about the distributional findings of Ganong and Shoag, who documented the end of convergence with a different method [4], or about Card, Rothstein and Yi's finding that housing costs consume the place premium for the average mover [7]. ◈ Strong Evidence Nor does it overturn the agglomeration literature, which places the productivity gain from doubling city size at three to eight percent [30]. What it undermines is the specific claim that housing restrictions in a few metros cost the United States a large share of its growth. The case for more housing in superstar metros may rest more on fairness and opportunity than on aggregate output.

Critics of the superstar framing raise further objections. Some argue that concentration is efficient and that the policy goal should be to help people move to opportunity rather than to bring opportunity to people. Others argue the reverse: that agglomeration benefits are overstated for most industries, that much of the urban premium is sorting [7], and that place-based policy can work where the foundations of a cluster exist [8]. A third camp, associated with Rodríguez-Pose, argues that neglecting lagging regions carries political costs that dwarf any efficiency loss [26]. ⚖ Contested The evidence assembled in this report gives partial support to each. The comparison below sets out the strongest version of the two main positions, the view of concentration as an engine of national prosperity and the view of concentration as a trap for those outside it.

Neither column is wrong. The engine case is strongest on productivity and on mobility for those already inside superstar metros; the trap case is strongest on access and distribution. The two can be reconciled by recognising that the same forces produce both outcomes. Agglomeration raises output [30], and housing scarcity determines who captures it [1]. If superstar metros built enough housing for the workers their multipliers create, the engine would operate as theory predicts and more workers would share in the premium [4]. ◈ Strong Evidence Because they do not, the engine runs primarily for incumbent homeowners and the highly educated. The policy question is therefore less whether concentration is good or bad than whether access to it is open or closed.

The debate also exposes a gap in the evidence. Most of the research cited here is American, because US administrative data are unusually rich. The extent to which Ganong and Shoag's convergence finding applies to London, Paris, Tokyo or Seoul is less well established [4], and the role of national institutions, from tax equalisation across regions to national wage bargaining and publicly owned housing, differs sharply between countries. The OECD's finding of slow regional productivity convergence across member countries [18] ◈ Strong Evidence and the ifo finding of declining district-level inequality in Germany [25] suggest that the American pattern is not universal. That is itself evidence that policy choices matter, and that the superstar outcome is a product of institutions as well as of economics.

The first conclusion is that housing supply in superstar metros remains the most direct lever on access, even after the Greaney correction [3]. The mechanism documented by Ganong and Shoag, in which migration toward high-wage places stops once housing absorbs the wage gain for lower-skilled workers, operates independently of any aggregate growth estimate [4]. ✓ Established Zoning reform that permits more homes near jobs in the most productive metros would let more workers share in the agglomeration premium and would weaken the capitalisation of that premium into land values [1]. The political obstacle is that the beneficiaries of such reform are mostly people who do not yet live in the metro and cannot vote there, while the costs, real or perceived, fall on incumbent residents who can.

The second conclusion is that place-based innovation policy can work only where foundations already exist, and must be concentrated rather than spread thinly. The Brookings growth-centres proposal argued for selecting a small number of interior metros with strong universities and existing technical workforces [8]. The US Economic Development Administration's Tech Hubs programme, created by the CHIPS and Science Act, authorised $10 billion for fiscal years 2023 to 2027, designated 31 regional hubs across 32 states and Puerto Rico in October 2023, and in July 2024 awarded $504 million in implementation grants to 12 of them, each between $19 million and $51 million [28]. ✓ Established The amounts are small relative to private venture investment of $339.4 billion in 2025 [29], and the results will take a decade to evaluate.

The third conclusion is that relocating government and paying people to move have limited power against market forces. Korea's Sejong City has absorbed 44 central government bodies since 2012, yet the Seoul capital area's share of the national population has passed 50 percent [23][22]. ◈ Strong Evidence Japan's relocation grant of one million yen per child has not prevented Tokyo's net inflow from rising again [31][21]. The UK's levelling-up missions were judged by the Institute for Government to be unlikely to reduce regional inequality as drafted [27]. The table below assesses the main policy instruments against the evidence assembled in this report, ranking each by the strength of the evidence that it changes who can reach the gains of the superstar metros.

The fourth conclusion concerns remote work. Its main geographic effect has been to hollow out downtowns and push housing pressure into suburbs and smaller places, not to move opportunity to lagging regions [13][14]. ◈ Strong Evidence Lagging regions that hoped to attract remote workers have found that those who came brought superstar salaries and, with them, superstar house-price pressure [14]. A remote-work strategy for lagging regions works best when paired with new housing supply in those regions, so that inflows add to the stock rather than bidding up the existing one. Without that pairing, the arrival of remote earners can reproduce in miniature the lock-out that defines the superstar metros, pricing local workers out of the towns they grew up in.

The fifth conclusion concerns measurement. The Hsieh-Moretti episode shows how a single model estimate can dominate policy for years [3]. The more durable evidence comes from convergent findings across independent datasets: house price divergence since 1940 [1], the end of income convergence after 1980 [4], the vanished non-college urban premium [5], the full offset of place premiums by housing costs [7], ✓ Established and the concentration of innovation and capital in a handful of metros [8][29]. Governments that track these indicators, rather than a single growth figure, will be better placed to judge whether their policies are opening access or merely moving it. Regional statistics offices in the UK, France, Japan and Korea already publish most of the necessary data [19][20][21][22].

The geography of opportunity in the 2020s is shaped by a paradox. The forces that make superstar cities productive are stronger than ever, amplified by the concentration of AI talent and venture capital [9][29], while the forces that once spread their gains, namely the migration of lower-skilled workers, the urban wage ladder and regional convergence, have weakened or stopped [4][6]. The evidence does not justify hostility to successful cities; their productivity funds much of what national governments do elsewhere. It does justify treating access to those cities as a central question of economic policy rather than a local planning matter. The superstar city is not the problem. The locked door is.