1.4.2 Regional Disparities

There are two Americas when it comes to AI. One is San Francisco, where 22.54% of the workforce is in tech, where AI job postings flood job boards, where venture capitalists deployed $22 billion into AI companies in 2024 alone—a 90% increase since 2020. Walk through the South of Market neighborhood and you pass OpenAI's offices, Anthropic's headquarters, dozens of AI startups occupying converted warehouses. The future is being built there, in real time, with enormous concentrations of capital.

The other America is rural Kentucky, where internet connectivity is unreliable, where "AI" is something people encounter on the news but rarely in daily life, where the factory that employed half the town closed years ago and nothing comparable has replaced it. Where the future is not being built—it is something happening elsewhere, to other people.

This is not an exclusively American story. Globally, 24.7% of the working-age population in the Global North uses AI tools, compared to 14.1% in the Global South. Africa holds less than 1% of global data capacity and would need an estimated $2.6 trillion in investment by 2030 to approach anything close to parity with high-income regions. AI is not spreading evenly. It is concentrating—in specific cities, specific countries, and specific regions—and everywhere it does not reach becomes relatively poorer, less productive, and more isolated from the economic gains the technology is generating. Understanding how this concentration works, and what sustains it, is essential to understanding the stakes of the AI era.

The Superstar Cities

The concentration of AI activity is striking in its specificity. San Francisco and San Jose alone accounted for approximately one-quarter of all AI conference papers, patents, and companies in 2021. Seattle claims the highest percentage of AI workers in the United States, with 20.81% of its workforce in artificial intelligence. San Francisco follows with 18.64% AI specialists among its tech workers, and Boston with 15.69% machine learning specialists.

This is part of a broader pattern of urban concentration in the innovation economy. More than 90% of U.S. innovation sector job creation between 2005 and 2017 occurred in just five major coastal cities, and that pattern has intensified rather than dispersed with the rise of AI. In 2025, AI companies attracted $3.3 trillion in investment—45% of all venture capital deployed that year—with the vast majority flowing to the Bay Area, New York, and Los Angeles.

The winner-takes-all logic of this geography is self-reinforcing. Capital flows to where capital already is. Talent clusters where talent already clusters. The presence of leading AI companies, research universities, and deep professional networks makes a city more attractive to the next wave of founders, engineers, and investors. Even other economically healthy cities struggle to compete. San Diego—a California city with world-class research universities and an established biotech sector—saw fundraising hit a six-year low in late 2025, with startups raising $897 million in the final quarter, down 36% from the prior year. The explanation offered by analysts was straightforward: venture capital was consolidating into the cities building AI, and San Diego had not yet secured a position among them.

The Fifty-Year Curse

History offers a sobering template for how AI concentration may unfold over the long term. Research on technology diffusion suggests that a major novel technology typically takes around fifty years to fully disperse geographically—and even after that dispersion, the pioneering locations retain dominance in the technology's highest-skill, highest-value activities for decades longer. Detroit remains the historic center of American automotive industry more than a century after the Model T. Silicon Valley still dominates semiconductor design half a century after the integrated circuit. Seattle has maintained its identity as an aerospace hub sixty years after Boeing's early rise to prominence.

The mechanism behind this persistence is well understood. Once a region establishes itself as a center for a new technology, it accumulates reinforcing advantages: specialized suppliers, a deep labor pool, research institutions oriented toward that sector, and a local culture of entrepreneurship that generates successive waves of startups. These agglomeration effects are genuinely difficult for competing regions to replicate after the fact. Latecomer regions can attract individual companies or research centers through subsidies, but recreating the ecosystem as a whole is a different and far harder challenge.

The implication for AI is significant. If high-skill AI activity is concentrating in 2025 and 2026 in San Francisco, Seattle, Boston, and New York, meaningful dispersion to second- and third-tier cities may not occur until the 2070s or beyond. Intermediate benefits—routine deployment of AI tools, AI-adjacent employment, productivity gains from off-the-shelf applications—may spread more quickly. But research centers, strategic decision-making roles, and the highest-paying positions tend to stay where they started. For policymakers and regional economic planners, this timeline demands urgent action rather than patience.

Urban Versus Rural

Within countries, AI's uneven spread manifests most clearly along the urban-rural divide. Urban workers in OECD countries are approximately 32% likely to be exposed to generative AI tools in their work, while rural workers face only a 21% likelihood—an eleven-percentage-point gap that represents millions of workers on fundamentally different productivity trajectories.

This disparity matters beyond the question of job displacement. AI exposure is also a driver of productivity gains, wage growth, and access to new economic opportunities. Workers who use AI to enhance their output become more productive over time; those who do not fall behind on a relative basis. As urban workers increasingly leverage AI tools in their daily work, they generate productivity growth that translates into higher wages and stronger local economies. Rural areas, where adoption is slower, see weaker wage gains and continued brain drain as skilled workers migrate toward cities where opportunities are more abundant.

The structural barriers to rural AI adoption are substantial and interconnected. Broadband internet remains unreliable or unavailable across large portions of rural America and similarly underserved regions worldwide. AI applications that depend on cloud computing require not just internet access but stable, high-bandwidth connectivity—a standard that remains aspirational in many rural communities. The universities and research institutions that produce AI talent are overwhelmingly located in metropolitan areas, and rural employers who want to use AI tools may struggle to find employees with the relevant skills locally. Each of these barriers compounds the others. Without deliberate policy intervention, the gap between urban and rural AI readiness grows wider each passing year.

The Global Divide

At the international level, the disparities are even more pronounced. Adoption rates in the Global North are growing nearly twice as fast as in the Global South. Internet access—the most basic prerequisite for any AI application—follows a steep gradient by income level: only 27% of the population in low-income countries has internet access, rising to 53% in lower-middle-income countries, 80% in upper-middle-income countries, and 93% in high-income countries. At every rung of this ladder, the gap in AI readiness between the bottom and top represents a real and growing constraint on economic opportunity.

Africa presents the starkest illustration of what these numbers mean in practice. A continent of 1.4 billion people holds less than 1% of global data capacity. Bridging that gap to approach parity by 2030 would require an estimated $2.6 trillion in investment—a sum that far exceeds the resources of African governments and for which no credible international funding mechanism currently exists. The consequences are concrete and physical: without data centers, latency makes real-time AI applications impractical. Without reliable electricity, AI systems cannot function at all. Without fiber optic cables and cellular infrastructure, none of the software benefits of AI can be accessed regardless of a country's policy ambitions. These are not software problems or regulatory problems; they are infrastructure problems of a magnitude that takes decades and enormous capital to address.

The Global South is not a monolith. Countries such as India, Brazil, and parts of Southeast Asia have rapidly growing AI sectors and are developing meaningful domestic AI capacity. But for the large majority of low- and lower-middle-income countries, the current trajectory points toward increasing marginalization from the AI frontier rather than convergence with it.

The Talent Drain

Regional disparities are further amplified by the migration of skilled workers toward AI opportunity. A computer science graduate from rural Iowa faces an employment landscape in which the highest-paying, most technically demanding AI positions are concentrated in a handful of coastal metropolitan areas. The rational individual response—and the cumulative result—is migration toward those hubs, draining human capital from the regions that can least afford to lose it.

The same dynamic operates at the global level. Engineers and researchers from Nigeria, the Philippines, India, and elsewhere who have the skills to contribute to AI development face strong incentives to relocate to the United States, the United Kingdom, Canada, or other high-income countries where AI companies and research institutions are concentrated. This brain drain is not a new phenomenon, but AI intensifies it by increasing the salary premium associated with technical skills and concentrating the most desirable positions in a small number of locations.

The cycle this creates is difficult to interrupt. Regions that lack AI activity lose their most skilled people to regions that have it. Their departure makes local capacity-building harder, reduces the likelihood that local firms will develop AI competencies, and signals to the next generation of students that their best opportunities lie elsewhere. The superstar cities, meanwhile, benefit from a continuous inflow of talent that reinforces their existing advantages. Reversing this dynamic requires either creating genuinely competitive local opportunities in lagging regions or substantially reducing the pull of the hubs—neither of which is straightforward without significant and sustained policy effort.

The Invisible Communities

One of the less-discussed dimensions of regional AI disparities involves populations that AI systems cannot adequately serve because they were never meaningfully represented in the training data. AI models are built on datasets that disproportionately reflect urban, high-income, and Global North populations. Rural and indigenous communities, speakers of minority languages, and populations in data-poor countries are often underrepresented or absent from these datasets, with consequences that extend well beyond missed economic opportunities. Tribal Nations in the United States are a striking domestic illustration — many reservation communities still lack reliable broadband, AI systems are almost never trained on data that reflects Native languages or health profiles, and the question of how tribes exercise sovereignty over their own data in the AI age has barely registered in mainstream governance conversations.

A medical AI trained primarily on data from hospitals in Boston and San Francisco may perform poorly for rural populations with different health profiles, different disease prevalences, and different patterns of healthcare utilization. A language model trained predominantly on English, Mandarin, and Spanish is of limited use to speakers of the world's 6,000-plus other languages. When these systems are nonetheless deployed in underserved contexts—often because they are the only available option or because their limitations are not fully appreciated by decision-makers—they can produce systematically worse outcomes for the populations that are least positioned to challenge or correct them.

This invisibility is structural rather than intentional. Data is collected where infrastructure exists, where technology companies operate, and where funding flows. Regions outside those zones become blind spots in the global AI training corpus, and their blind-spot status reduces the incentive to invest in data collection there, which perpetuates the blind spot. Breaking this cycle requires deliberate effort: targeted data collection initiatives in underrepresented communities, investment in local AI research capacity, and governance frameworks explicitly designed to account for the needs of populations that market forces alone will not serve.

The Policy Vacuum

Fewer than one-third of developing countries have national AI strategies, and approximately 118 nations—the majority from the developing world—remain absent from the major multilateral forums where AI governance is being shaped. The practical consequence is that the global rules governing AI development, deployment, and accountability are being written largely by and for high-income countries, with limited input from governments representing the majority of the world's people.

This exclusion has concrete implications for how AI systems are built and regulated. Standards and regulatory frameworks developed in high-income contexts embed assumptions about infrastructure, institutional capacity, data protection, and use-case priorities that may not translate to lower-income settings. When AI systems designed to these standards are deployed in developing countries—sometimes by international development organizations promoting them as tools for poverty reduction—they may fail to deliver promised benefits, introduce new forms of bias, or extract economic value without generating commensurate local benefit.

The barriers to meaningful participation in AI governance are not primarily matters of political will. Developing countries often lack the financial resources to fund delegations to international forums, the technical staff to engage substantively with complex regulatory proposals, and the institutional infrastructure to implement governance standards even when they agree with them in principle. Addressing the policy vacuum requires more than extending formal invitations to the table—it requires resourcing the capacity to participate effectively, which means funding technical assistance, training, and institution-building in lower-income countries over the long term.

What Concentration Costs

The economic costs of regional AI concentration are substantial, though difficult to measure precisely. The most direct cost is unrealized productivity growth: regions that cannot adopt AI technologies miss the efficiency gains that would otherwise translate into higher output, higher wages, and improved living standards. Beyond this, there are innovation costs—potential AI applications that are never discovered because the problems they might solve are not visible to the AI researchers and developers clustered in a handful of high-income cities. And there are human capital costs: people in lagging regions who have the potential to contribute to AI development but never gain access to the education, mentorship, or opportunity to do so.

At the macro level, Brookings Institution research suggests that geographic gaps in AI readiness risk producing long-term economic divergence, in which pioneering regions compound their advantages over time while lagging regions fall into structural traps—conditions in which slower AI adoption leads to weaker growth, which leads to reduced infrastructure investment, which makes AI adoption even harder. This divergence dynamic is not hypothetical; it is the pattern that characterized the digital economy over the past two decades, and there is little in the current trajectory of AI development to suggest the outcome will be different without intervention.

The political costs are equally significant. When entire regions observe economic opportunity flowing to distant cities while their own communities stagnate, the result is resentment and political instability. The deindustrialization of manufacturing regions in the United States and Europe during the 2000s and 2010s offers a recent illustration: the economic displacement was real, the political backlash was real, and the policy response proved inadequate to contain either. AI is creating the conditions for a similar dynamic—potentially at greater speed, and affecting a wider range of workers and communities.

Pathways to Convergence

The concentration of AI activity is not an immutable feature of the technology—it is a product of policy choices, investment decisions, and institutional arrangements that could, in principle, be made differently. The analytical challenge is not identifying what kinds of interventions would help; the policy toolkit is reasonably well understood. The challenge is whether the political will and resources exist to deploy them at the necessary scale.

Domestically, effective interventions would include substantial investment in rural and underserved broadband infrastructure, public funding for AI research centers in second- and third-tier cities, tax incentives for companies that locate AI operations outside established hubs, and expanded access to AI education and workforce training in lagging regions. The United States has historical precedents for this kind of deliberate geographic redistribution: the rural electrification programs of the 1930s, the interstate highway system, and the land-grant university network all expanded economic opportunity across a broader range of communities. Analogous programs oriented toward digital infrastructure and AI-focused regional development could pursue similar goals, though they would need to be sustained over decades to match the scale of the challenge.

Internationally, bridging the Global North-South AI divide requires a more ambitious agenda: substantially increased funding for digital infrastructure in low-income countries, technology transfer mechanisms that allow developing countries to build domestic AI capacity rather than simply purchasing AI services from foreign companies, and genuine inclusion of developing-country voices in AI governance. Existing multilateral institutions have acknowledged these needs in principle, but concrete commitments have remained well below the scale of the problem. The $2.6 trillion investment gap in African data infrastructure alone illustrates the mismatch between the stated ambitions of international development actors and the actual resources being mobilized.

What makes convergence difficult is not ignorance of the solutions. It is that the current concentration benefits the actors with the most power to shape investment and policy: AI companies, venture capital firms, and governments in high-income countries. Redistributive policies require those actors to accept constraints on advantages they currently enjoy—which historically has required either sustained political pressure from disadvantaged groups or a shared recognition that rising inequality poses systemic risks that ultimately harm everyone, including those who appear to be winning.

Summary

AI's geographic concentration operates simultaneously at multiple scales and is one of the defining features of the current moment in technological development. Within countries, a small number of major metropolitan areas—led by San Francisco, Seattle, Boston, and New York—capture the overwhelming majority of AI investment, talent, and innovation. Globally, the divide between the Global North and the Global South mirrors and amplifies these domestic inequalities, with internet access, data infrastructure, and AI adoption rates varying dramatically by national income level.

Several interconnected mechanisms sustain this concentration. Agglomeration effects make existing hubs more attractive to the next wave of capital and talent, while historical evidence suggests that the highest-value activities of a new technology take approximately fifty years to disperse geographically. Infrastructure deficits—in broadband, data centers, and reliable power—prevent adoption in rural and lower-income regions even where the motivation to adopt exists. Talent migration drains human capital from lagging regions toward established hubs. Underrepresentation in training data renders many communities effectively invisible to AI systems, degrading the quality of AI-driven services they receive. And the absence of developing-country participation in AI governance means that the global regulatory framework is being designed without adequate input from the majority of the world's population.

The costs of this concentration are measurable in unrealized productivity, foregone innovation, and deepening political instability. The interventions capable of reducing it—infrastructure investment, inclusive regional development policy, international technology transfer, and governance reform—are known, but have not been pursued at the required scale. Without a significant change in trajectory, the geographic inequalities already visible in AI development are likely to widen, with consequences that will extend well beyond economics into politics, social cohesion, and the long-term distribution of power.

Key Takeaways

  • AI investment is concentrating in a handful of cities: San Francisco and San Jose alone account for roughly a quarter of global AI papers, patents, and companies; the Bay Area, New York, and LA captured 45% of all venture capital in 2025.
  • Historical evidence suggests high-skill AI activity concentrated now may not meaningfully disperse to second-tier cities for roughly 50 years — agglomeration effects (specialized suppliers, talent pools, research institutions) are self-reinforcing and extremely difficult to replicate after the fact.
  • Urban-rural gaps are significant and compounding: urban OECD workers are 32% likely to use generative AI in their work; rural workers are 21% — an 11-point gap translating into diverging productivity, wages, and economic trajectories over time.
  • Global disparities are more severe: only 27% of low-income country populations have internet access (versus 93% in high-income countries), and Africa would need an estimated $2.6 trillion to approach parity in data infrastructure — far exceeding any existing international funding mechanism.
  • The talent drain reinforces concentration: skilled workers from lagging regions migrate to AI hubs, reducing local capacity-building and signaling to the next generation that their best opportunities lie elsewhere.
  • Underrepresented communities are invisible to AI systems trained on data skewed toward urban, high-income, Global North populations — degrading the quality of AI services they receive and reducing investment incentives to collect data there.
  • Fewer than one-third of developing countries have national AI strategies, and 118 nations are absent from major multilateral AI governance forums — meaning the global rules of AI are being written without meaningful input from the majority of the world's population.

Sources:

Last updated: 2026-03-02