Some of the greatest advances in technology have emerged from bringing intelligent people together to solve problems. How do tech clusters develop & how can we use them to replicate past successes?
When an industry has thus chosen a locality for itself, it is likely to stay there long: so great are the advantages which people following the same skilled trade get from near neighbourhood to one another. The mysteries of the trade become no mysteries; but are as it were in the air, and children learn many of them unconsciously.
Alfred Marshall
As technological advances continue to build out the digital economy it is perhaps ironic that physical locations still seem to matter so much. Slack and Twitter and Zoom and a host of various collaboration software products have facilitated our ability to communicate with people from all around the world. And yet, up-and-coming founders will still pay 3x the national average in rent so that they can live in Silicon Valley or Manhattan. Despite our ever-improving virtual tools, the triumph of the city is not reversing anytime soon. And the countries with the most vibrant tech clusters will get to decide the pace and the direction of the future. We need both a better understanding of how tech clusters form and how we can maximize their benefits.
Why and how do tech clusters form?
Economists and urbanists talk a lot about the significance of clustered networks of individuals and firms working and living together in a shared environment. All else equal, having more smart people interacting leads to more ideas which leads to more innovation which leads to more growth. The whole ecosystem of product designers, scientists, engineers, supply chain managers, investors, and academics ends up creating more than the sum of their parts. These agglomeration effects occur in cities across the world, but the US has had a unique geopolitical advantage in having the premier industrial clusters for essential technologies like software development and machine learning.
Defined formally, we could say that an industrial or tech cluster is a geographic region with a disproportionate share of economic activity, high-skilled technical employment, granted patents, and R&D funding when compared to their share of the national population. Less formally, we could say that tech clusters are in the cities where the most interesting conversations and cutting-edge applications in a particular field are happening.
| Consolidated Metro Area | Venture Capital Investment | Granted Patents | Employment in Top 10 R&D industries, highly skilled | Population |
| Boston | 10.5% | 4.5% | 5.5% | 1.6% |
| Los Angeles | 6.5% | 5.3% | 5.6% | 5.8% |
| New York | 15.3% | 6.0% | 6.3% | 6.4% |
| San Francisco | 48.1% | 18.4% | 11.7% | 2.5% |
| Seattle | 2.1% | 4.0% | 4.2% | 1.2% |
San Francisco is the archetypal example, but Boston, Seattle, Austin, and Denver all meet this threshold as well. Larger cities like New York and Los Angeles are more ambiguous, as their share compared to population size is not disproportionate, but narrowing the analysis to a subregion like Manhattan or Santa Monica would likely turn up a traditional tech cluster.
Historically, clusters have been pivotal in driving long-term US growth and for creating innovations that improve the lives of billions of people around the globe. As economists William Kerr and Frederic Robert-Nicoud summarize, there has been a continual movement of leading tech clusters over time in the US. In the 1800s, Lowell, Massachusetts was the center for textile mills relying on water power. By the early 1900s, Cleveland, Ohio was instrumental in pushing forward the frontier on electricity and steel. Detroit, Michigan, of course, developed into the powerhouse for automobile manufacturing in the mid-1900s.
Currently, US tech clusters are the envy of the world. There are only four trillion dollar companies in the world. Two of them are based near San Francisco (Apple and Alphabet), and two near Seattle (Amazon and Microsoft). Of the global top 30 Internet firms, 14 are based in SF alone.
The effects that these clusters have on innovation and productivity growth are quite impressive, even on the individual level. Enrico Moretti models their impact: “a computer scientist moving from the median cluster in computer science (Gainesville, FL) to the cluster at the 75th percentile of size (Richmond, VA) would experience a 12.0% increase in productivity, holding constant the inventor and the firm. In biology and chemistry, a move from the median cluster—Boise, ID—to the 75th percentile cluster—State College, PA—is associated with a productivity gain of 8.4%, holding constant the inventor and the firm.” And as the clusters get larger and more specialized, the productivity boosts also get larger.
While the benefits and significance of tech clusters are fairly well established, what is less well known is precisely how tech clusters originate and why they end up in the specific places they do. That software clustered near San Francisco was not inevitable, that auto manufacturing clustered in Detroit was not fate, that semiconductor fabrication today clusters in Taiwan is not preordained. So how do clusters arise?
In the past, clusters would frequently populate around a specific natural resource that was necessary for the operation of the tech in question, i.e. natural harbors, coal and iron deposits, fast-flowing streams. But the ultimate resources for technology development today are the human minds that can come up with and then implement new ideas. So tech clusters today are mostly about attracting and organizing talented humans in cities they would actually want to live in.
More formally, to become a tech cluster today, a location needs:
- An anchor organization as a first-mover in getting high-skilled talent in a particular field to come to a region (usually a firm or a university)
- An urban environment in which firms and workers in adjacent sectors can benefit from large knowledge spillovers as new advances and technical approaches diffuse across the region quickly
- The development of high-velocity labor markets where workers with deep technical expertise can quickly start, join, or leave various firms
- A social scene that is animated by a particular industry or technology area such that the conversations happening in local bars are some of the best in the world for that subfield
- A sufficient supply of industry-specific resources (venture capitalists, relevant scientific or manufacturing infrastructure, government contracts)
- A good bit of luck
It’s worth emphasizing that the development of a cluster is a process rather than a flip that is switched at a specific moment in time. A location may not have all of these elements right at the very beginning but will need to quickly develop them if it wants to have staying power beyond an initial flash in the pan. Also complicating things, some of these factors suffer from a chicken-or-the-egg problem as it is much easier to attract specialized capital and interested young people to a region once the cluster already exists.
The historical accidents of particular government policies can sometimes serve as a tiebreaker among competing destinations. One reason that Silicon Valley may have become the primary driver for software development rather than the Boston area is California’s decision to not enforce non-compete agreements. While not necessarily the intention, this accelerated both the degree of knowledge spillovers and the velocity of labor markets in the region. Today, top developers at large tech companies frequently leave to start competitors and help cutting-edge practices and the more promising technical approaches diffuse quickly throughout the region. This may be frustrating for the particular firms losing out, but it makes the larger tech cluster more vibrant and dynamic. And by increasing the number of approaches attempted to solve an issue, it increases the odds of a breakthrough innovation happening.
But tech clusters don’t necessarily last forever. As an industry matures and becomes more established it tends to be less reliant on the kind of rapid experimentation and ideation that is facilitated so well by tech clusters. As that happens, an industry may disperse over a larger set of regions, or be subsumed into a more traditional urban agglomeration.
Unsurprisingly, it seems to be the case that the development of a new general-purpose technology is usually a prime candidate for allowing new tech clusters to form. But interestingly, there may be a bias towards slightly smaller cities that can operate as more a blank slate for the fledgling industry than can a large, established city.
Looking at the emergence of auto manufacturing as an example, despite having fewer initial entrants and a smaller population, activity started to concentrate in smaller cities like St. Louis, Cleveland, Indianapolis, and in the ultimate winner, Detroit. According to Kerr and Robert-Nicoud, this smaller city advantage may have been due to the higher physical proximity of engineers, production line managers, and funders as it was easier to share prototypes, run experiments, and circulate ideas between the relevant stakeholders. These smaller regional hubs may have been able to provide more attention and financial resources to these new firms because there was simply less competition.
This should not be all too surprising when we consider the softer, cultural aspects that make up an attractive tech cluster. It seems likely that the culture of a city made up disproportionately of geneticists and CRISPR scientists might differ from that of petroleum engineers and fracking specialists. Having a separate city is useful for establishing a different culture and offers a type of specialization that can attract a separate set of weird, ambitious young people that will form the backbone of a cluster.
The conversations in the bars and social scenes of San Francisco feel very different from those in the financial district of NYC because the types of people attracted to those cities are very different. To truly become a cluster for a particular field, a city essentially has to reach some critical density of conversations in the social scene to be ones happening between like-minded specialists. That is naturally much easier to do in a smaller city with less preexisting focus. This may help explain why we don’t see one giant mega-city dominate all-new tech cluster development.
Will tech clusters survive remote work?
All of the promising digital tools created by these clustered tech firms present an interesting counter-scenario for tech clusters. During the present COVID-19 crisis, a number of large tech firms, including Twitter, Facebook, and Quora, have made the move to fully remote work. Similarly, some companies like Stripe have offered one-time moving bonuses (and small pay cuts) to their employees if they actually take the leap out of Silicon Valley. This has led some to speculate that the future of remote work is upon us and tech clusters will slowly (or perhaps quickly) erode.
While it’s unclear to what extent these policies will be reversed after the eventual distribution of a COVID-19 vaccine, it does represent a fascinating (and perhaps worrisome) natural experiment. The internal effect of innovation should be internalized by the companies making these decisions (are Facebook engineers creating as many good ideas for Facebook), but they are not internalizing the impact this might have on public spillovers and new firm creation. Put differently, will the number of firms created by big tech employees exiting their company be severely diminished if all their interactions with potential co-founders and funders are being mediated via Zoom? Some have argued that they will not be. But I am inclined to worry that they will.
The bull case for remote work is simple. There are many restrictions on the current operation of tech clusters (more on this later) that can be arbitraged away by shifting jobs across city or even national borders. Immigration and housing restrictions end up kneecapping the ability of tech firms to bring in the most talented up-and-coming workers and from housing them at affordable rates. Remote work offers a way around this. Furthermore, digital agglomeration effects may be starting to approximate physical agglomeration effects in some ways. If you want fascinating in-depth conversations in your field, you can find great ones on Twitter or Reddit.
However, certain kinds of human interactions will be very difficult to replicate. The spontaneous interaction with a colleague you hadn’t thought of, the friendship that is easier to forge working late hours in the office than over a zoom chat, the brainstorming session that happens more naturally in person than over a laggy connection, etc.
It is also worth pointing out that “digital vs physical interactions” is the wrong way to think about this tradeoff. After all, in tech clusters today, workers get the best of both worlds. We get the best leads and conversations the internet has to offer combined with the spontaneous interactions, dense physical networks, and close access to specialized resources of the physical space. So a world with widespread remote work will still benefit from tech clusters so long as physical human interaction provides benefits to innovation above and beyond the best interactions offered digitally. This is especially true for the superstar firms and workers for which even small boosts in productivity and idea generation can have far-reaching consequences.
Part of this staying power will also come from assortative mating. Smart, creative people generally like to be friends with and date other smart, creative people in the same or adjacent industries. Then, as those power couples go on to look for jobs elsewhere, they are more likely to consider cities that have deep enough job markets for both spouses. So long as friendship, dating, and marriage markets continue to rely on some degree of physical proximity it seems likely that these clusters will persist. This in turn creates a situation where employers want to be located near these deep labor markets of technical talent.
If software development, specifically, does end up being distributed across the country and less reliant on strong clustering effects, that may be an indication that the field has become more mature and/or less complex. Think of the move from website development as a highly technical field that required manually coding everything from scratch to the platform-based system we see today in which hundreds-of-thousands of design firms around the country can cook something up quickly using WordPress. There are already some early signs that basic software development could head that direction as new AI tools like GPT-3 may be able to do most of the hard work.
But, of course, that would partially be an indicator that the intellectual energy has shifted into other fields (like AI!) that are more interesting and will be the focus of attention for tomorrow’s tech clusters.
Likewise, San Francisco as a specific cluster could be on the decline as decades of bad housing policy are finally catching up with it, and remote work does make an exodus of talent easier than it once was. But tech clusters as a concept and as an economic force to be reckoned with are not going away.
Policy for tech clusters
Given the importance of tech clusters for driving innovation and the fact that they aren’t going to decrease in importance anytime soon, it’s worth considering policy tools that we could leverage both to strengthen existing tech clusters and to increase the likelihood of new ones developing.
To strengthen existing tech clusters, the fix may be surprisingly simple to identify. Policymakers can use immigration reform to give them access to talented people from all over the world, use zoning reform to give them plentiful housing and conducive urban environments in which to live and work, and give local universities enough research funding to supply the public good of basic science development which undergirds their work.
Of course, the implementation of those details is devilishly difficult. Congressional attempts to reform our immigration system have stalled for decades. Meanwhile, Canada’s tech clusters like Toronto are thriving from the talented students and entrepreneurs that have been turned away by the US. There are interesting executive reforms to the immigration system using tools like the Schedule A occupation list, International entrepreneur rule, and O-1 visa for extraordinary ability that deserve more attention and that could serve as a stopgap until Congress passes comprehensive immigration reform.
Despite the critical importance of the Silicon Valley cluster for US international competitiveness, a self-inflicted housing shortage has kept out thousands or potentially even millions of individuals who could be better contributing to US innovation and productivity growth. While particularly acute in San Francisco, this problem is widespread across urban centers in the US. This creates a geographical mismatch between workers and the clusters they would be best suited for, which in aggregate reduced US growth by 36% between 1964 – 2009, according to research from Chang-Tai Hsieh and Enrico Moretti. Figuring out ways to politically enable the creation of dense, walkable cities in which inventors, academics, supply chain managers, engineers, and designers can live, work, and overlap is a difficult but essential task then.
Federal science funding, meanwhile, has been slowly declining as a share of GDP since the 1960s, despite it’s important role in facilitating private sector investment. The easy answer would just be to simply supercharge existing science funding institutions like the National Science Foundation and DARPA. But the structure and incentives in science funding are far from ideal, and may be actively counterproductive in ways that simply scaling the existing system could make outcomes worse. Instead, we need to pair increased investment with reforms to improve the grantmaking process, scale back the bureaucracy, increase research diversity, and figure out what actually works.
Trying to seed new tech clusters may be even more tricky, especially from the top down. Recent history is littered with failure after failure in attempts to recreate the magic of Silicon Valley elsewhere. Attempted Silicon Valley knock-offs including Silicon Spuds in Idaho, Silicon Hill in DC, Silicon Bayou in Louisiana, and many, many others. Very few of which have had any success.
It’s simply not going to work if we arbitrarily decide that Cleveland is the next hub for genetic editing and then spend billions trying to make that happen. But that doesn’t mean we are powerless to affect the creation or spread of tech clusters. Rather than being primarily focused on the outcome of achieving a tech cluster in location X, we should spend more time thinking about and trying to seed the preconditions for tech clusters so that they might arise “naturally”. As soon as early indications start looking good (even if due to random chance), additional research funding and university commercialization efforts can turn on an investment spigot and act as miracle growth for a fledgling sector.
The key here is some degree of industry neutrality and competition. Rather than funding a specific technology or company, policymakers should provide a general flow of research funding in a topic area. Obviously individual grants have to be directed to specific companies or ideas, but in aggregate, a portfolio approach makes sense and maximizes the chance of a big breakthrough. Universities are one way to funnel these funds in a firm-neutral way, and allowing individual professors to play a role in driving commercialization efforts is helpful for keeping the agenda connected to the cutting-edge scientific work. Trying to recreate some of the magic of corporate R&D labs from the 1960s and 70s is another interesting approach that is beginning to see more attention. Ultimately, funding from a variety of sources and towards institutions that have very different sets of incentives (academic vs corporate vs public labs) seems like a promising path forward.
As an additional area for future research, I think we need a clearer understanding of what impact local university patenting policies play in incentivizing fruitful commercialization efforts. Since the Bayh-Dole Act in 1980, universities have been able to benefit from the IP created from federally-funded research projects. But policies around what cut individual professors can get vary significantly from university to university and could perhaps be better optimized.
As long as there are scientific and technological advancements to be made, progress on the frontier will inevitably cluster in some physical locations. Until we humans are plugging our brains directly into computers, we remain social animals who need physical interaction to best stimulate, develop, and actively work out new ideas. The highest digital worlds we can create will rely on the careful curation of our physical one. Pretending otherwise is to delay the difficult policy work that needs to be done to start maximizing our tech clusters instead of actively undermining them as we’ve been doing for decades.