Network Effects

Network effects exist when a product becomes more valuable to each user as more users join, creating a self-reinforcing dynamic where leaving becomes increasingly costly. They are the strongest category of moat available to a startup because they compound rather than depreciate. They explain why a handful of platforms (Facebook, Uber, eBay, LinkedIn, Visa, Microsoft Excel) dominate their categories despite having no patent or proprietary technology that competitors could not replicate.

The major types: direct network effects (one-sided, also called same-side) where each user benefits directly from more users of the same kind, as in phones, fax machines, WhatsApp, or Zoom; two-sided or multi-sided network effects where multiple distinct user types benefit from each other's presence, as in marketplaces (eBay's buyers and sellers, Uber's riders and drivers, Airbnb's guests and hosts) or platforms (iOS's users and developers, YouTube's viewers and creators); data network effects where the product itself improves as more users contribute data, as in Google Search (each click teaches the ranking algorithm), Waze (each driver improves traffic data), or modern LLM-powered products that learn from user interactions; and social network effects where the existing relationship graph creates lock-in beyond the product's utility, as in Facebook, LinkedIn, or Snapchat. The classic mathematical framing is Metcalfe's Law: the value of a communication network scales with the square of its users (n²), so doubling users roughly quadruples value. Real-world network effects rarely hit n² in pure form; Reed's Law (group-forming networks scale as 2^n) and Sarnoff's Law (broadcast networks scale linearly with n) bracket the realistic range. The strongest moats stack multiple types: Uber has direct network effects (more riders means shorter wait times via density), two-sided effects (more drivers means more riders and vice versa), and data network effects (better routing and pricing from more rides). The defensibility test that matters: when a competitor offers your users a better product, does the network make leaving cost them something that the competitor cannot replicate? If yes, you have real network effects. If no, you have growth or virality, not a moat.

What founders get wrong: Confusing virality, growth, and network effects. Virality is a customer-acquisition mechanism (existing users bring new users). Growth is the rate of user addition. Network effects are a defensibility property (existing users would lose something if the network shrank). A product can be viral with no network effects (most consumer apps), and can have network effects without virality (most B2B marketplaces). They are independent properties; do not conflate them.

Related: Moat · Defensibility · Viral Coefficient · Data Flywheel · Product-Led Growth

FAQ

What's the difference between network effects and virality?
Virality is how users acquire new users (referral mechanics). Network effects are why users stay (value increases with network size). A viral product spreads fast but may not retain. A product with strong network effects retains because leaving costs users their existing connections, data, or transactions.

What are the main types of network effects?
Four main types: direct (one-sided, like phones), two-sided (marketplaces, like Uber), data (the product improves with more user data, like Google Search), and social (relationship graphs create lock-in, like Facebook). Strongest moats stack multiple types.

Is Metcalfe's Law accurate for real businesses?
Approximately. Metcalfe's Law (value scales as n²) is a useful ceiling but rarely holds exactly. Real networks face local clustering, churn, and quality decay that flatten the curve. Most real network effects scale somewhere between n log n and n². The point is that growth compounds, not that the equation is precise.