Often, the most educational part of a bubble narrative is the one that didn't come true. Bubbles happen when we start extrapolating, and they work when those extrapolations are mutually-reinforcing, but navigating them means being able to think about many different parties' behavior, many steps ahead. Dot-com boosters were right that we'd move some of our shopping and a lot of our media consumption online (though at least in the 90s, they underestimated how much socializing would move online). But brick-and-mortar retail didn't die, and while the value of the news business was impaired by unbundling classifieds from local news and by creating endless niche audiences for national news, both businesses still exist in some form. And one thing that definitely didn't happen was that free access to information, and the ability to transmit it instantly, would make physical distance less relevant. You can start a software company anywhere, but a large share of them are still within a one-hour drive of Stanford. And while a hedge fund can be located anywhere with an Internet connection, they're still very likely to be a short walk from a Metro North stop.

In fact, these places seem to be more central than they used to be. For physical businesses, colocation often makes a lot of sense because it reduces the cost of transporting products around—that effect will be especially powerful for small purchases, which can speed up the development of prototypes by capital-constrained new entrants to an industry.¹ But for purely information-driven businesses, it seems intuitive that they'd be spread-out, and that someone who didn't like New York's weather or California's tax policies could sit in an identical office chair typing the same commands into an identical computer in whatever location made the most sense to them.

And even if those ties were held together through force of habit, perhaps some big disruption could have jostled everything into a new configuration. If, for example, there were a pandemic that made in-person interactions some combination of a health risk and a legal one, and people scattered to the suburb or survivalist compound of their choice, it would be hard to bring back the network effect.

But, if anything, Covid made those effects stronger. The roster of second-tier cities shifted a bit, and Chicago in particular lost some high-profile residents, but New York and the Bay Area both seem as dominant as ever. Frontier AI, for example, seems more physically concentrated in the Bay Area than SaaS or consumer Internet was.² Given unlimited options, people chose the previous default, only more so.

One reason for this is that all of those time- or space-agnostic communications tools—Zoom calls, emails, blog posts that turn out to be a message to someone Googling that exact topic years later, etc.—are complements to in-person interaction, which remains high-bandwidth. It's easier to meet someone remotely, but still hard to actually get to know them. Moreover, it's easier to get together a critical mass of people with some interest so niche that few cities will be big enough to support regular meetups for it.

If you spend all of your time reading about and chatting about AI online, you have a simulacrum of what it would look like if everyone in the world were obsessed with AI. Pick the right group house in Berkeley and the right day job, and you can basically live in that world 24/7. Digitally-mediated experiences can compete with the real thing on some axes, or even beat it—the real world offers fewer opportunities than video games for global conquest, shooting aliens, etc.—but it's not the best substitute for high-value interactions.

It's also harder for purely digital environments to engineer the right level of serendipity. The basic setup for many recommendation filters is to show a pretty random/standard mix of content to new users, and then to rapidly guess what their narrow interests are based on their behavior and that of similar users who've produced more data. But it's still artificial, and it also means that the fringiest ideas never get a chance. Over time, the strongest gravitational pull is exerted by revenue, and the platforms that generate the most revenue per user have the biggest budget for providing services tailored to those users. The result is that the online experience is a strange hybrid of a quirky downtown superimposed on an endless digital suburbia. There isn't a good data-driven way to deal with this, especially because the impact of serendipity is both predictable and laggy. But cities seem to provide the right mix, and finding the right spot on an abstract digital map of social relations turns into lead generation for finding the physical place to be.

The Diff has covered city economics and their contrast with the digital in contexts that range from conference recaps to mid-pandemic speculation about what the post-Covid future will hold:

• The Future of Remote is Not That Remote

• Covid Baby Bust or Baby Boom? ($)

• Conferences as Industry Cluster Economics, Squared

• The Bloomberg Terminal Shows How to Create Virtual Economic Cluster

Join the discussion!

1  In other words, you don't want your future Henry Ford to be burning cash on overhead while he's waiting for a part from Buffalo. Especially if it turns out to be the wrong one to order, after all. If there's a dense network, it makes intermediate goods more available and more predictable, and as anyone who's ever worked on or modeled a complex multi-step process knows, it only takes a few high-variance steps to make the average pace unpredictable. The same intuition that explains why the best place to start a car company was the one with the most competitors explains why "dark kitchens" have short menus.

2  This might be a side effect of the fact that many corporate nerve centers were located in New York, making it a good place to have a physical presence if part of the job is convincing Fortune 500 CIOs to adopt something new. And, of course, social and search are ad products, and the ad buyers and sellers tend to congregate in NYC. Trading begets trading, once again!