Moat types
Atomic networks: win a small market before you try to win a big one
Updated 21 October 2026
An atomic network is the smallest group of users that can make a networked product work on its own and keep growing: one team, one campus, one city, one niche. The term comes from Andrew Chen's book on the cold start problem. Startups that win a durable network almost always win an atomic network first.
The idea is simple. A product that depends on other users is useless when it is empty. So instead of launching everywhere at once, you pick a group small enough to fill completely, make the product work for them, then copy the pattern to the next group.
The cold start problem
Andrew Chen, a general partner at a16z who previously worked at Uber, describes the cold start problem plainly: "you have a phase in all these products where you don't have any users so the product isn't working." His book, The Cold Start Problem, draws on interviews with teams at LinkedIn, Zoom, Dropbox, Tinder, Uber and Airbnb.
In an early essay on social products, Chen argued that density beats raw size: you are better off with 100,000 users who each have 30 connections than 1,000,000 users who each have 2. A big, thin launch inflates sign-ups but leaves each user with an empty product, so they churn.
The atomic network is the answer to that problem. As Chen put it: "If you can build one atomic network, you can build a second and a third."
Why it matters for startups, not big companies
Big companies can launch a networked product to an existing user base. Google, Meta or Microsoft can switch a feature on for hundreds of millions of people and skip the cold start. A startup cannot. It has no installed base, little money for paid acquisition and one shot at a first impression.
That asymmetry is also where the advantage lies. A large company tends to launch broadly and thinly. A startup can go deep on one small group the incumbent ignores, and become the default there before anyone else notices. Peter Thiel makes the same argument in Zero to One: "Every startup should start with a very small market. Always err on the side of starting too small."
Winning a small market is not the moat itself. It is how the moat starts. Once a startup owns many dense local networks, a later entrant has to win each of them back one at a time, which is what makes network effects defensible.
Types of starting markets
The right atomic network depends on how the product creates value. Common shapes:
- A team or company. Workplace tools start inside one team, then spread across the organization and to its partners. Slack and Yammer fit here.
- A campus or school. Students live close together, talk constantly and share a calendar. Facebook, Tinder and Snapchat started this way.
- A city or neighborhood. Marketplaces with physical supply need local density: drivers near riders, couriers near restaurants. Uber, DoorDash, Yelp and Nextdoor started here.
- A niche community. A group with a shared interest or event, such as Twitter's early tech crowd at SXSW.
- A narrow customer segment. For B2B, Geoffrey Moore's beachhead market means "targeting a single, highly challenging use case in a single industry and just nails it." His rule of thumb is to pick targets "big enough to matter but small enough to lead."
Atomic network examples
| Company | First market | What it shows |
|---|---|---|
| Harvard students | A tiny market that felt built for its users | |
| Slack | 6 to 10 friendly companies | Teams as the unit of adoption |
| Zoom | 2 to 3 people on a call | Some atomic networks are very small |
| Tinder | USC campus | Seed the most connected users at once |
| Uber | San Francisco | Win one city, then copy the playbook |
| DoorDash | Palo Alto | Manual service in one town before software |
| Nextdoor | One Menlo Park neighborhood | A pre-connected group adopts fast |
| Airbnb | New York hosts | Go door to door in one dense market |
- Facebook. Paul Graham notes it was at first "just for Harvard students," a market of a few thousand people, "but because they felt it was really for them, a critical mass of them signed up."
- Slack. Before its August 2013 preview, the team begged and cajoled friends at 6 to 10 companies, including Rdio, to use it. When Rdio expanded from a small group to its 120-person engineering team, Slack learned how the product behaved at a larger scale.
- Zoom and Slack sizing. Chen has said Zoom's atomic network is two to three people, while Slack's is five to ten using it at the same time.
- Tinder. According to Chen's account, the first version already had the right features but not the right network. A party at USC that required the app for entry put it on the phones of about 500 well-connected students at once, and Tinder repeated the play at other campuses.
- Uber. It went live in San Francisco in May 2010 and did not launch in New York until May 2011. One city had to work before the next.
- DoorDash. It began as Palo Alto Delivery, a landing page with PDF menus of local restaurants and the founders' phone number. They delivered the first orders themselves.
- Nextdoor. Its pilot in summer 2010 was Lorelei Manor in Menlo Park, about 90 households with an established homeowners' association. The company later expanded to around 140,000 neighborhoods.
- Airbnb. Graham describes the founders "going door to door in New York, recruiting new users and helping existing ones improve their listings."
- Yelp, Twitter, Snapchat, Yammer. Chen lists local saturation examples: Yelp in San Francisco, Twitter in the SXSW tech community, Snapchat in Southern California high schools and Yammer within companies.
- Reddit. The founders seeded the site with many accounts they controlled to set the tone of the community until real users took over. It is a reminder that the first network sometimes has to be built by hand.
- Clubhouse (cautionary). Downloads fell from 9.2 million in February 2021 to roughly 900,000 in April, according to Sensor Tower data. Huge top-of-funnel interest does not replace dense groups of users who keep coming back for each other.
How a startup builds an atomic network from zero
- Define the unit of value. Ask how many people, and which ones, must be present for one user to have a good experience. For a chat tool it may be a team of five. For a ride marketplace it is enough drivers in one area to keep wait times short.
- Pick a market you can fill completely. Choose the smallest group where you can realistically reach most of the people who matter: one company, one dorm, one neighborhood, one specialty.
- Find the hard side. Most networks have a side that is harder to attract, such as drivers, hosts, creators or sellers. Recruit it by hand first.
- Seed everyone at once. Density matters more than totals. Tinder's party and Slack's team rollouts worked because a whole group arrived together.
- Do the unscalable work. DoorDash delivered food itself. Airbnb went door to door. Manual effort is acceptable while the network is small.
- Measure local health, not global totals. Track retention, activity and liquidity inside the first market. If that market is not self-sustaining, launching a second one will not help.
- Copy and expand to adjacent markets. Move to the next campus, the next city, the next team inside the same customer. Expansion should reuse the playbook and, ideally, the existing network.
A starting market also strengthens other moats. A product that owns a community from day one often gains a distribution advantage that competitors cannot buy.
Common mistakes and fake versions
- "Our first market is everyone." A launch aimed at all small businesses or all consumers is not a starting market. It is a large, thin network.
- Starting markets that are too big. "New York" may be too broad for a hyperlocal product. A borough, a building type or a single industry may be the real unit.
- Counting sign-ups instead of density. Large waitlists and press coverage can hide the fact that no single group is getting value.
- "Network effects at scale." If value only appears with millions of users and there is no small group that benefits on its own, the startup has no realistic path through the cold start.
- Picking a market nobody can expand from. A great first market should lead somewhere. A niche with no adjacent markets caps the business.
- Confusing a pilot customer with a network. One enterprise customer using a tool alone is a customer. It becomes an atomic network only when users inside it depend on each other.
How Moat Check scores it
Moat Check calls this test Winnable starting market. It asks one question: "Is there a small, self-contained market it can dominate first (a team, a school, a city, a niche)?" The reasoning, from the methodology: durable networks start by dominating a small, self-contained group before expanding.
- Pass (10 points): A clearly defined first market (a team, campus, city or niche) that can be won on its own.
- Partial (5 points): A starting market is plausible but not clearly defined, too broad to win, or depends on something not yet in place. Claims that the product works "at scale" score a partial at best.
- Fail (0 points): Value only appears at large scale, with no realistic way to start small.
The test pairs closely with copycat resistance. A rival with more money can copy features quickly, but taking over a market where you already own the dense network is much harder. Moat Check uses AI and web research to apply this fixed framework, so treat the score as a starting point for your own thinking, not a verdict or investment advice. For the full picture, see the other moat tests or read what a moat is.
Questions
What is an atomic network?
An atomic network is the smallest group of users that can sustain a networked product on its own, such as one team, one campus or one city. Andrew Chen popularized the term in The Cold Start Problem.
What is the cold start problem?
The cold start problem is the early phase of a networked product when there are too few users for it to be useful, so new users leave. Solving it usually means building one dense, self-sustaining network before expanding.
How big is an atomic network?
It depends on the product. Andrew Chen has described it as two to three people for Zoom, five to ten for Slack, and roughly 500 people on a college campus for Tinder.
What is the difference between an atomic network and a beachhead market?
A beachhead market, from Geoffrey Moore's Crossing the Chasm, is a narrow customer segment you dominate first in order to expand. An atomic network is the network version of the same idea: the smallest group of connected users that makes the product work.
Is an atomic network a moat?
Not on its own. It is how a defensible network starts, because a rival must win each small market one at a time. The moat comes from owning many dense networks that a later entrant would have to rebuild.
Keep reading
Who will copy your startup?
Real competitors, the likely copycat and a moat score. Free, about a minute.