Why skilled people cluster
Modified
Input
Skilled people cluster where employers, colleagues, knowledge, capital, and professional mobility reinforce one another. High salaries alone rarely create a durable talent center; workers evaluate expected learning, future options, institutional quality, and household costs. A region attracts specialized AI talent by building a productive system in selected domains, not by declaring a general ambition to become a hub.
The Geography of Capability
Digital work appears to weaken geography. Code, papers, data, and meetings can cross borders at low cost. Yet specialized talent remains strongly clustered.
AI researchers, quantitative analysts, semiconductor engineers, biotechnology scientists, and creative professionals often concentrate in a limited number of regions. They do so not only because firms are there. Firms also locate there because talent, suppliers, investors, universities, and specialized knowledge are there.
The result is cumulative causation: concentration produces advantages that attract further concentration.
A Location Choice Model
Let worker $i$ choose location $r$ to maximize expected utility
where:
- $w_{ir}$ is expected compensation;
- $L_{ir}$ is learning and knowledge access;
- $O_{ir}$ is the value of future career options;
- $A_{ir}$ is amenities and institutional quality;
- $C_{ir}$ is housing, tax, relocation, and household cost;
- $R_{ir}$ is regulatory, cultural, and professional risk; and
- $\varepsilon_{ir}$ captures idiosyncratic preference.
A salary premium can be offset by high living costs, weak professional mobility, limited schools, visa uncertainty, or low trust in institutions.
Talent attraction is therefore a bundle.
Agglomeration Economies
Clustering raises productivity through several mechanisms.
Table 1. Sources of the productivity premium from clustering
| Mechanism | How it operates | AI example |
|---|---|---|
| Labor pooling | Firms and workers face lower matching risk | A specialist can change employers without leaving the region |
| Knowledge spillovers | Ideas move through seminars, projects, and informal contact | Researchers learn which methods are failing before publication |
| Specialized suppliers | Niche services become locally viable | Data infrastructure, legal, compute, and evaluation providers |
| Capital access | Investors learn to assess domain-specific risks | Technical diligence becomes more credible |
| Reputation | The location itself attracts customers and collaborators | External organizations search the cluster first |
These mechanisms create external returns. A firm's presence can raise the productivity of nearby firms even without a formal transaction.
A Production Function with Local Spillovers
Suppose firm $f$ in region $r$ produces
where $\overline{H}_r$ is the region's average relevant human capital and $D_r$ is the density of complementary organizations.
If $\phi,\psi>0$, the firm gains from the surrounding cluster.
The private firm does not own these terms. It benefits from universities that train people, competitors that deepen the labor market, professional associations that spread standards, and suppliers that reduce fixed costs.
This creates a policy rationale for shared infrastructure, but also a warning: subsidizing one visible firm is not equivalent to producing $\overline{H}_r$ or $D_r$.
The Option Value of a Thick Labor Market
Relocation is risky for a specialist. A job may disappoint, a project may fail, or a firm may close.
In a thick labor market with many potential employers, the worker retains an outside option. Let $p_r$ be the probability that the first job ends and $V_r^{next}$ the expected value of the next local opportunity. The location's career value is
A region with one attractive employer may offer a high current wage but a low $V_r^{next}$. The worker and household must price the risk of another relocation.
This explains why isolated institutions struggle to attract senior talent. The candidate evaluates the ecosystem, not just the offer.
Learning as a Location Premium
Early- and mid-career specialists may accept lower current compensation for faster learning.
Let future skill evolve as
where $\rho_r$ measures the region's learning environment.
Working with capable colleagues, difficult problems, research seminars, and demanding clients can raise the future wage path. The present value of the location includes
A region that offers a salary but weak learning may attract short-term labor without retaining ambitious specialists.
Why General “Hub” Strategies Fail
Governments and institutions often announce ambitions to become an AI hub. The category is too broad.
AI capability is embedded in domains: finance, manufacturing, medicine, energy, logistics, public administration, science, or creative work. Each domain requires different data, regulation, capital, customers, and professional communities.
Read as a diagnostic rather than a second scorecard, the framework becomes clear: Promote AI as a universal sector — Domain-centered strategy: Select problems with existing local capability. Build generic office or compute capacity — Domain-centered strategy: Build infrastructure tied to actual workloads. Recruit prominent individuals — Domain-centered strategy: Develop teams, suppliers, and career options. Count events and memoranda — Domain-centered strategy: Measure repeated production and knowledge spillovers. Offer temporary subsidies — Domain-centered strategy: Reduce durable institutional and matching frictions.
A credible cluster begins where the region already has some combination of demand, knowledge, and productive organizations.
Attraction and Retention Are Different
A recruitment package can bring a person to a region. Retention depends on daily experience.
Let the probability of remaining after period $t$ be
where $B_i$ is household belonging and $F_i$ is institutional friction.
Friction includes uncertain residence status, administrative opacity, language barriers, exclusion from decision making, weak research support, or limited opportunity for a spouse.
If these conditions are poor, recruitment spending subsidizes circulation rather than accumulation.
The Role of Universities
Universities can anchor a cluster when they perform more than credential production.
They can:
- train specialists to a verifiable standard;
- host seminars through which tacit knowledge circulates;
- maintain shared research infrastructure;
- connect external firms to faculty and students;
- support experiments too uncertain for one firm;
- create public technical materials; and
- attract visitors who widen the network.
The university's contribution is strongest when research, teaching, and external problems form a cumulative system. A campus isolated from firms and professional communities produces fewer spillovers.
Remote Work and Partial Clustering
Remote work changes the geography of production without making location irrelevant.
A distributed team can separate residential location from organizational location. Let collaboration quality be
where $\omega$ is the share of interaction conducted remotely.
Remote systems work well when tasks are modular, documentation is strong, and performance is observable. Local interaction may remain valuable when tasks are ambiguous, knowledge is tacit, or trust must be built rapidly.
Hybrid clusters can therefore emerge: a small institutional core connected to a globally distributed production network. Their success depends on deliberate coordination, not on video calls alone.
Digital connection can also widen a cluster without dissolving its center. A seminar may include participants from several countries, a research repository may be public, and a distributed firm may recruit internationally. The core still supplies recurring relationships, standards, and the people who decide which problems deserve attention. Remote participants gain access to part of the cluster's knowledge system, while the cluster gains a larger search space for ideas and collaborators.
This suggests a useful distinction between access and membership. Digital tools can make information accessible at low cost. Membership in a productive community usually requires repeated contribution, feedback, and reputational exposure. A cluster that publishes its work and maintains international interfaces can attract contributors before they relocate. It can then reserve physical concentration for activities whose tacit or institutional content justifies the cost.
A Strategy for a Specialized Cluster
A region or institution should begin with a narrow capability map:
- Which industries already produce internationally credible output?
- Which specialized skills are scarce relative to real demand?
- Which universities or firms can train and evaluate those skills?
- What prevents specialists from accepting or remaining in roles?
- Which shared assets would lower costs for several organizations?
- How will the cluster remain connected to global knowledge?
The strategy becomes credible when specialists can imagine not one job, but a professional future.
That future must also remain open to newcomers. If a cluster protects incumbents, recycles the same relationships, or treats outside criticism as disloyalty, density stops producing learning. Healthy concentration combines strong local ties with porous intellectual boundaries.
Remote work changes the boundary of a cluster without eliminating its logic. A specialist can serve an employer abroad, but still benefits from living near peers, research institutions, investors, and alternative employers. Some knowledge can be codified and transmitted cheaply; other knowledge travels through repeated interaction, observation, and trust. The result is partial unbundling: the firm, worker, client, and knowledge community no longer need to occupy the same place, yet the worker may still pay a location premium to remain inside a dense learning market. Policies built only around attracting headquarters can therefore miss the more important objective of sustaining professional communities in which skill continues to compound. Cluster policy should accordingly track repeated collaboration, specialist job mobility, research-to-firm links, and the survival of new ventures. A branded district or subsidized building is an input; the network of learning and matching is the productive outcome. The distinction matters for evaluation. Occupancy can rise while knowledge exchange remains weak, and a smaller community can generate high spillovers when its members repeatedly solve problems, move between institutions, and finance one another’s experiments. Clusters should be judged by these repeated productive relationships rather than by the number of organizations whose addresses fall inside a designated boundary. A genuine cluster continues producing these interactions after the original subsidy, event, or prominent employer has disappeared.
Recent AI investment illustrates the geography of complements. The Economy’s analysis of AI exposure geography argues that places differ in their capacity to turn technological change into productive opportunity. The AI Trade Shock Will Reward Countries That Move Workers, Not Just Data adds labor mobility and adjustment, while the data-center jobs analysis shows that headline facilities often conceal the wider network of construction, energy, maintenance, and specialized services. Clusters form around that network, not around a single celebrated asset.
Conclusion
Skilled people cluster because productivity, learning, matching, and future options are geographically interdependent.
High compensation can attract an individual, but a durable talent center requires capable colleagues, multiple employers, knowledge institutions, specialized suppliers, capital, and livable conditions. These complements create the external returns that make a cluster cumulative.
For AI, specialization matters. A region rarely becomes globally attractive by pursuing “AI” in the abstract. It becomes attractive by solving difficult problems in domains where local knowledge, demand, and institutions can support productive careers.
The unit of talent policy is therefore not the vacancy. It is the ecosystem around the vacancy.
References
Edward L. Glaeser, Hedi D. Kallal, José A. Scheinkman, and Andrei Shleifer, “Growth in Cities”, Journal of Political Economy, 1992.
Enrico Moretti, The New Geography of Jobs, Houghton Mifflin Harcourt, 2012.
William R. Kerr, The Gift of Global Talent, Stanford University Press, 2018.
The Economy Editorial Board (2026) ‘AI Exposure Geography Is a Capacity Problem, Not a Red-Blue Problem’, The Economy Review, 11 June.
The Economy Editorial Board (2026) ‘The AI Trade Shock Will Reward Countries That Move Workers, Not Just Data’, The Economy Review, 2 June.
The Economy Editorial Board (2026) ‘The AI Data Center Jobs Debate Is Counting the Wrong Workers’, The Economy Review, 17 August.
Similar Post
