One City, 60,000 Cards: The Density Rule Behind Every Network That Works
Emrah · In 1958, Bank of America launched its first credit card in a single city instead of the whole country. Andrew Chen's idea of the atomic network explains why that was the right call, and why founders and investors read the exact same growth chart so differently.
The Story
In 1958, Bank of America wanted to launch the first mass-market credit card in the United States. The obvious move, given the size of the opportunity, would have been a national rollout: hit every major city at once, get the card into as many wallets as possible, as fast as possible.
Instead, the bank mailed 60,000 unsolicited credit cards to households in a single city: Fresno, California. Not a test market alongside a bigger launch. The entire initial push, concentrated in one place.
This story is the anchor for Chapter 6 of Andrew Chen's book The Cold Start Problem: How to Start and Scale Network Effects (2021), titled "The Atomic Network — Credit Cards," part of the book's "Cold Start" section spanning chapters 4 through 10. Chen is a general partner at Andreessen Horowitz who spent years before that on Uber's growth team, and the book draws on interviews with founders and early team members from Slack, Tinder, Zoom, Twitch, Reddit, Uber, Airbnb, and PayPal, among others, to build a framework for how network-effect businesses actually get off the ground.
The Fresno decision looks conservative on the surface. It was the opposite. A credit card is a two-sided product: it is worthless to a cardholder if merchants nearby do not accept it, and worthless to a merchant if too few customers carry one. Launching thin across many cities would have produced a card that almost nobody could actually use anywhere. Concentrating the entire launch in one city meant Bank of America could get both sides, cardholders and merchants, dense enough in that one place for the card to become genuinely useful to somebody.
The Idea
Chen's term for this minimum viable density is the atomic network: the smallest network that can sustain itself without artificial life support, whether that support is marketing spend, subsidies, or manual sales effort. Below that size, nothing works no matter how much money you throw at it. Above it, the network starts feeding itself.
What counts as "enough" varies enormously by product. Two people are enough for a video call to matter, which is why Zoom's atomic network is essentially trivial. One team, often under ten people, was enough to make Slack sticky inside a company, after which it could spread team by team. Airbnb needed hundreds of active listings within a single market before supply and demand could reliably find each other. A credit card needed the merchant and cardholder density of an entire city.
The mechanism underneath all of these examples is the same two-sided dynamic: value on one side depends on sufficient presence on the other side, in the same place, at the same time. Spread a network thin across many markets and this loop never fires anywhere, because density never crosses the threshold in any single location. Concentrate it in one place and the loop fires once, and then the network can sustain and expand itself from there. This is also why the atomic network breaks down in two specific ways: when a team underestimates the actual threshold (declaring victory at 50 listings when the real number is closer to 500), or when they draw the boundary too wide (a country instead of a city, an entire company instead of one team), diluting density below the point where anything self-sustains.
How Founders Should Read This
For founders building anything with network effects, the practical discipline is to make the initial launch smaller and narrower than instinct suggests. Uber's early growth playbook is a clean example: rather than targeting "a city," early expansion efforts focused on absurdly specific pockets of demand, famously described as something like 5pm at the Caltrain station at 5th and King Street in San Francisco. Once rides were reliably available at that specific place and time, the radius expanded outward from a proven core, rather than starting broad and hoping density would emerge on its own.
The trap runs in the opposite direction, and it is a common one. Picture a marketplace founder six months into building a platform, eager to show investors that the business is scaling. Instead of concentrating effort, the team launches in eight cities simultaneously, picking up 40 to 50 listings in each. The total looks respectable on a slide: 400 listings. But if the real atomic threshold for that category of marketplace is a few hundred active listings within a single city, then none of the eight markets has actually crossed it. No city has enough supply to reliably satisfy demand, so no city generates the self-reinforcing loop that turns early listings into a real marketplace. The better version of that same six months would have put all 400 listings into one city and made that single market genuinely dense before expanding at all.
The uncomfortable part of this discipline is that it usually means saying no. No to the second city that seems ready. No to the adjacent vertical that looks like low-hanging fruit. No to the enterprise customer who wants a feature built for their specific, non-representative use case. Every one of those yeses dilutes the resources that would otherwise go toward finishing the first atomic network.
How Investors Should Read This
Investors evaluating the same growth chart are, or should be, asking a structurally different question than "how many users do you have." The more useful version is: which single market or segment has already crossed into self-sustaining density, and what is the actual number behind that threshold?
This reframing exposes a common red flag: a company that is live in many markets or segments but has not clearly crossed the density threshold in any single one of them. Total user counts can climb steadily in this scenario while the underlying business is not actually getting stronger, because breadth without density does not compound. It can look like traction on a dashboard while representing, in reality, many small, disconnected experiments rather than one working network.
Fund economics is the reason this distinction matters so much to a venture investor specifically. Returns at venture scale depend on a small number of portfolio companies becoming very large, which in turn depends on a repeatable expansion motion: Uber going city by city, Slack going team by team, Airbnb going market by market. An investor is effectively underwriting the bet that whatever worked to build the first atomic network can be copied and pasted into the next one, and the one after that. A network that required a completely bespoke, one-off strategy to reach critical mass in its first market, with no clear template for the second, is a weaker bet on this dimension even if that first market is thriving. Ownership targets and reserve decisions for follow-on rounds get built on top of this replicability assumption, which is exactly why the density question matters more to an investor than the aggregate user count does.
Where the Two Views Collide
The founder's natural instinct, particularly heading into a fundraise, is to look broad. Being live in twenty cities or three countries reads as momentum to journalists, to potential customers, and often to less experienced investors. It is a understandable instinct, and not a dishonest one. Breadth is a real, visible signal of effort and ambition.
The investor's job is to look past that signal to what it is actually built on. Breadth without density is frequently closer to a mirage than to evidence: it can mask the fact that no single market has genuinely tipped into self-sustaining growth. The founder's read of the situation, that visible width equals traction, and the investor's read, that width without depth proves very little, can both be internally consistent and still point to opposite conclusions about the same set of numbers.
The person who understands this distinction has an advantage at the table, regardless of which side of it they sit on. Reframing the conversation away from total user counts and toward "which single market or segment has crossed the atomic threshold today, and what is that number" turns a vague traction story into a specific, falsifiable claim. A founder who can answer that question with a real number is showing something stronger than a growth chart. An investor who knows to ask it is not being difficult. They are asking which of those 400 listings actually did any work.
The Takeaway
Density before breadth is not a slogan, it is a testable claim about a specific number in a specific place. Before the next market, the next vertical, or the next feature aimed at a new segment, the more useful exercise is naming the current atomic network, if one exists yet, and the number that defines its threshold.
Which single segment of your current users, customers, or markets is dense enough today to keep growing without you actively pushing it, and what number proves that?
Source: Andrew Chen, "The Cold Start Problem: How to Start and Scale Network Effects," 2021 — coldstart.com
