The Paradox of the Expensive Address
You're sitting in a conference room on the forty-third floor of a Manhattan tower, watching the rent bill in your head tick upward by the minute, and you are thinking: why are we here? The firm across the street charges more per square foot than some small nations collect in tax revenue. The talent costs a fortune. A blizzard can shut the whole operation down for days. Silicon Valley sits on an active fault line, charges rents that would make a London landlord wince, and experiences periodic wildfires. Los Angeles has no natural harbor, a chronic water problem, and gridlock so severe that a "quick meeting" can consume an entire morning. And yet global finance clusters in lower Manhattan, semiconductor design clusters in Santa Clara County, and the machinery of the entertainment industry concentrates, stubbornly, around a desert basin in Southern California.
The geographic disadvantages are real. The clustering happens anyway.
The short answer is that geography is almost never why an industry starts in a place, and once it's there, almost nothing can move it. The longer answer involves a set of interlocking mechanisms that economists call agglomeration effects, and understanding them changes how you think about cities, competition, and why your company almost certainly cannot relocate to a cheaper zip code and keep everything else intact.
The Accident That Became a Moat
Most clusters begin with something embarrassingly contingent. The film industry landed in Los Angeles partly because independent producers in the early twentieth century were fleeing the reach of Thomas Edison's Motion Picture Patents Company, which enforced its equipment patents aggressively on the East Coast. California was far enough away. The weather allowed year-round outdoor shooting, and the landscape was varied. Those practical advantages have long since evaporated: indoor lighting solved the weather problem decades ago. And yet the industry stayed, because by the time those original reasons became irrelevant, something more durable had replaced them.
That something is what economists call thick labor markets. When enough firms in one industry occupy the same city, they collectively create a pool of specialized workers that no single firm could justify training on its own. A mid-sized visual effects studio in Los Angeles can hire a rigging supervisor, a Houdini pipeline technical director, or a creature effects specialist within weeks, because hundreds of other productions have already trained those people. The same studio in Indianapolis would have to import those workers or pay to build them from scratch. The labor market in Los Angeles is thick precisely because everyone else is already there, which keeps everyone there.
This is the core of the trap, and it runs in both directions. Workers cluster where the jobs are. Jobs cluster where the workers are. Each new entrant makes the pool slightly deeper for everyone else, including their direct competitors.
Knowledge Spills Across the Street, Not Across the Country
Here is the part most explainers skip, and it is the one that actually matters. Proximity accelerates the transfer of tacit knowledge: the kind that doesn't travel well in documents or video calls.
Tacit knowledge is what you learn by watching someone debug a problem, by overhearing a conversation at lunch, by being in the room when a deal goes sideways and seeing how the senior people handle it. It is the stuff that doesn't fit in a manual. Research going back to Alfred Marshall's observations about English industrial districts in the 1890s consistently finds that this kind of knowledge moves through geographic proximity far more effectively than through formal channels. Two engineers who used to work at the same semiconductor fab, now at rival firms two miles apart, will exchange more useful information at a conference happy hour than they would in a year of LinkedIn messages. The cluster is, in this sense, a slow-motion school that never closes and never charges tuition.
Consider two product designers. One works at a furniture startup in Milan, surrounded by leather suppliers, metalworkers, and competitors who have been refining the same craft for generations. The other takes the nominally identical job at a firm in a mid-sized city where the industry has no history. Both have the same formal training. Within three years, the one in Milan will have absorbed an education that no curriculum delivers, purely through proximity and osmosis. The cluster produces knowledge that the cluster then keeps.
This is why the venture capital industry concentrated so heavily in a few streets in Menlo Park, even as the internet supposedly made geography irrelevant. The deals that get done there depend on reputation, trust built over shared failures, and pattern recognition that comes from watching hundreds of companies up close. That knowledge is not in any database.
The Supplier Ecosystem Nobody Talks About
Clusters also generate something less glamorous but equally sticky: deep local supply chains. The fashion industry in northern Italy isn't just designers and brands. It's a dense network of specialized fabric mills, button manufacturers, pattern-cutting services, and dyeing facilities concentrated in a relatively small radius. A designer in Milan can spec a fabric, get a sample, revise it twice, and have the finished bolt in hand faster than a competitor in London can get a single quote.
This supplier density is self-reinforcing in the same way labor markets are. Specialized suppliers locate near their customers. Their presence attracts more firms that need those suppliers. Each new firm makes the ecosystem more complete. The whole thing becomes extraordinarily hard to replicate from scratch somewhere else, because you would need to convince hundreds of interdependent businesses to move simultaneously.
Nobody does that.
The catch: this same density can make a cluster brittle. When Detroit's automotive cluster began contracting, the supplier network contracted with it, making it harder for any remaining manufacturer to operate there, which accelerated the contraction further. Agglomeration effects run in reverse, too, and a thick market thinning out can thin faster than anyone expects.
What People Get Wrong About This
The popular assumption is that remote work and digital infrastructure have dissolved these dynamics. They haven't, at least not for the industries where tacit knowledge and trust matter most. Ask yourself: when was the last time a genuinely transformative deal, in finance, in film, in venture capital, was closed by people who had never been in the same room? What remote work has actually done is make the periphery of a cluster more viable. You can now be a screenwriter in Austin and maintain genuine connections to the Los Angeles industry. But the core, the green-light decisions, the packaging of projects, the relationships that determine whose calls get returned, that core remains physically concentrated. The cluster's gravitational center hasn't moved; its radius has expanded slightly.
The other thing people get wrong is treating a city's geographic disadvantages as evidence that something is broken. They are usually just irrelevant. The fog and cold of San Francisco did not stop the software industry from concentrating there. The variable that mattered was the presence of Stanford, then the presence of early firms, then the presence of the people those firms trained, then the presence of the investors who funded the next generation, and so on until the cluster became self-sustaining. Geography, in the end, is just where the first domino happened to fall.
Once those dominoes are standing in a line, the question of whether the location is optimal becomes almost academic. The real question is whether the concentration of knowledge, workers, suppliers, and trust in that specific place outweighs the friction of leaving. For most mature clusters, it does, by a considerable margin. The conversation about why some industry doesn't just move somewhere cheaper tends to end the same way every time: with the industry still exactly where it was, and the question quietly dropped.