Network effects: the structural moat that separates the largest software companies from the rest.

What network effects actually are

A network effect exists when a product becomes more valuable to each user as more users join. The classic example is the telephone: one phone has zero value; two phones have some value; a billion phones have enormous value because any user can reach any other user. Network effects are the structural moat that separates the largest software companies (Facebook, LinkedIn, Airbnb, Uber, Slack) from competitors with similar products and worse network density. Andrew Chen wrote the definitive operator-level treatment in The Cold Start Problem (2021).

The mechanism is non-linear. Metcalfe's Law says the value of a network grows as the square of the number of users. With 10 users, there are 45 possible connections. With 1,000 users, there are about 500,000 connections. The value-per-user rises with network size, which makes the leader's lead structurally hard to close. The second-place network can be 30 percent behind in users but 50 percent behind in value-per-user.

Andrew Chen distinguishes between direct network effects (users add value for other users directly, like Facebook), indirect network effects (more buyers attract more sellers and vice versa, like Airbnb or eBay), and data network effects (more users produce more data that improves the product, like Google Search or recommendation engines).

Claim: Andrew Chen's The Cold Start Problem (2021) argues that all network-effect products go through five stages of development: the cold start problem, the tipping point, escape velocity, hitting the ceiling, and the moat. Each stage requires different growth tactics. Source: Andrew Chen, The Cold Start Problem (Harper Business, 2021). Context: Chen wrote the book after a decade at Uber, Andreessen Horowitz, and various network-effect product companies. The five-stage framework is now standard vocabulary for operators working on marketplaces, social networks, and collaborative tools.

Three types of network effects

Three patterns cover almost every network-effect product. Direct network effects (each user adds value for other users directly). Indirect or two-sided network effects (one side attracts the other, like Airbnb's hosts attracting guests). Data network effects (more users produce data that improves the product). Each type behaves differently in the cold-start phase and has different scale-stage dynamics.

The three main network effect types with examples and key dynamics

Type	How it works	Examples
Direct	Each new user makes the product more valuable for existing users	Facebook (more friends to connect with), LinkedIn (more professionals), Slack (more team chatter)
Indirect / Two-sided	More buyers attract more sellers; more sellers attract more buyers	Airbnb (hosts and guests), Uber (drivers and riders), eBay (sellers and buyers)
Data	More users produce more data that improves the product for everyone	Google Search (queries improve results), Waze (driver data improves routing), Spotify (listening data improves recommendations)

The cold start problem

Network-effect products face a chicken-and-egg problem at launch. The product has no value until enough users join, but users will not join a product that has no value. Andrew Chen's term for this is the cold start problem. Solving it requires either picking a tightly-defined niche where the network has value at small scale, or running expensive supply-side and demand-side acquisition simultaneously until the network reaches its tipping point.

The niche strategy is what most successful network-effect products used. Facebook launched at Harvard, then a handful of universities, before expanding. The network had value with just hundreds of users because they were the same students who would otherwise see each other on campus. The niche let the network density matter at small scale.

The simultaneous-acquisition strategy is more expensive but sometimes necessary. Uber and Airbnb both ran heavy supply-side and demand-side acquisition in launch markets, often subsidizing both sides to bootstrap density. The cost was substantial but the resulting network created moats competitors could not match without similar investment.

How to design for network effects

Four design principles separate products that build network effects from products that look networked but never compound. Pick a niche where the network has value early. Make the first session valuable before the network is built. Design product mechanics that strengthen with use. Measure network density, not just user count.

Niche first. Pick a vertical, geography, or use case where the network has value at small scale. Facebook started with Harvard. Yelp started in San Francisco. Slack started inside a single gaming startup. Wide nationwide launches almost always fail for network-effect products.

First-session value. The first user should get value before any other users join. Slack's first single-user session has value (file storage, search, notes). Uber's first ride works even if there are only three drivers. Products that require a built network to deliver any value cannot escape the cold start.

Mechanics that strengthen with use. The longer a user is on the product, the more value they extract and the more inputs they create. Slack's mechanic strengthens as messages accumulate (history, search, context). Pinterest's strengthens as users save more pins (recommendation quality improves).

Measure density, not size. The right metric is messages per active user per week, transactions per user per month, or matches per user per quarter. Total user count masks whether the network is actually working. A million users with low density loses to 100,000 users with high density every time.

Quick answers

What is a network effect in plain English?

A product gets more valuable to each user as more users join. One telephone has zero value. A billion telephones have enormous value. The same dynamic powers Facebook, Airbnb, Uber, and most large software companies.

What is Metcalfe's Law?

The value of a network grows as the square of the number of users. 10 users = 45 possible connections; 1,000 users = ~500,000 connections. Non-linear scaling makes leader's leads structurally hard to close.

What is the cold start problem?

Andrew Chen's term for the chicken-and-egg problem at the launch of a network-effect product. The product has no value until enough users join, but users will not join a product without value.

How did Facebook solve cold start?

Niche launch. Facebook started at Harvard, then a handful of universities, before expanding. The network had value with just hundreds of users because they were the same students who would otherwise see each other on campus.

What are the three types of network effects?

Direct (Facebook, LinkedIn). Indirect or two-sided (Airbnb, Uber). Data (Google Search, Spotify). Each type has different cold-start dynamics.

What should I measure for a network-effect product?

Network density, not user count. Messages per active user per week. Transactions per user per month. A million low-density users loses to 100,000 high-density users every time.

Frequently asked

What is a network effect?

A product or service that becomes more valuable to each user as more users join. The telephone, fax machine, social networks, and marketplaces all show this dynamic. Network effects are the structural moat behind the largest software companies in 2026.

Who studies network effects?

Andrew Chen wrote the definitive operator-level treatment in The Cold Start Problem (2021). Hal Varian wrote earlier academic work. James Currier at NFX has published extensively on the taxonomy of network effects.

What is the cold start problem?

Andrew Chen's term for the chicken-and-egg problem at launch. The product has no value until enough users join, but users will not join a product with no value. Solving cold start requires niche launches or expensive simultaneous supply-and-demand acquisition.

What are direct network effects?

Each user adds value for other users directly. Facebook (more friends to connect with), LinkedIn (more professionals to network with), Slack (more colleagues to message), Telegram (more contacts in the app).

What are indirect or two-sided network effects?

Two sides where each attracts the other. Airbnb (more hosts attract more guests; more guests attract more hosts). Uber (drivers and riders). eBay (sellers and buyers). The dynamic is harder to bootstrap than direct effects because both sides have to grow together.

What are data network effects?

More users produce data that improves the product for everyone. Google Search (more queries improve relevance). Waze (more drivers improve routing). Spotify (more listening improves recommendations). Often combined with other effect types.

How long does it take to build a network?

Years typically. Facebook took roughly 18 months to escape Harvard. Airbnb took 4-5 years to reach tipping point. Most network-effect products fail because they run out of capital before reaching the tipping point.

Can a company without network effects compete with one that has them?

Sometimes. Slack competed with Microsoft Lync and Cisco WebEx by being a better product before the network effect was built. Once Slack reached density, the network effect locked customers in. The competitive window is narrow but real.