The engine that makes winners unbeatable

Most products are worth the same to you regardless of how many other people own one. A hammer is as useful whether ten people or ten million own the same model. Its value lives in the object.

A small class of products breaks this rule: they get more valuable to each user as more people use them. A telephone is useless if you are the only owner and indispensable when everyone has one. The value is not in the handset; it is in the network of other users the handset connects you to.

This property is a network effect, and it is the single most important idea in platform economics. It explains why some markets end up dominated by one or two giants while others stay fragmented forever, why those giants are so hard to dislodge, and why the same companies keep winning category after category.

The whole of this path unpacks that one idea: where network effects come from, how a platform ignites them despite a brutal cold start, and why, contrary to the usual story, they do not always produce a monopoly.

A network effect is a feedback loop, and drawing it out shows why it is so powerful.

more users -> more value per user -> product more attractive
     ^                                        |
     |________________ attracts more users ___|

Each turn of the loop makes the next turn easier. More users make the product more valuable, which attracts more users, which makes it more valuable still. This is the flywheel: hard to start, because early on there are few users and therefore little value, but increasingly hard to stop once spinning.

The strategic consequence is asymmetry over time. At the start the flywheel works against you: with few users the product is genuinely worse, so growth is a grind. Past a threshold it flips and works for you: the product's value now grows on its own as users arrive, and growth compounds.

This is why platform businesses look so different from ordinary ones. A normal product is roughly as good on day one as at scale. A network-effect product is worse when small and better when large, by its own nature, which makes the early period existential and the late period nearly unassailable.

The simplest kind: the product gets better as more people just like you use the same product. Users on one side, all benefiting from each other.

Communication tools are the pure case. A messaging app is valuable in proportion to how many of the people you want to talk to are on it. A social network is worth joining when the people you care about are already there. In each, users benefit directly from other users of the same type.

This produces the strongest lock-in of all, because the thing keeping you is not the software, which a rival can copy in months, but the people, whom a rival cannot copy at all. You might prefer a competing app's features and still not switch, because your friends are not there. The switching cost is social, not technical.

It also explains why direct-network-effect markets tip so hard toward one winner. If everyone benefits from being where everyone else is, everyone converges on the same place, and a second-best network with half the people is often worth far less than half as much, so it struggles to survive at all.

The subtler and often more powerful kind involves two different groups who each make the product more valuable to the other, without directly benefiting from their own kind.

A payment card is the classic example. Cardholders do not benefit from other cardholders. But more cardholders attract more merchants who accept the card, and more accepting merchants attract more cardholders. Each side pulls the other. The value crosses between groups rather than within one.

The same shape appears everywhere platforms sit between two groups: an operating system with users and app developers, a marketplace with buyers and sellers, a ride app with riders and drivers, a console with gamers and studios. In each, growth on one side is a magnet for the other.

Indirect effects are the foundation of most large platform businesses, and they are harder to bootstrap, because you must get two groups to show up, and each is waiting for the other. That chicken-and-egg problem is severe enough to be the entire subject of the next lesson. For now the key point is the structure: value flows between sides, so a platform's job is to be the place where both sides expect the other to be.

A more modern kind, and the one most tangled up with confusion: the product improves because more usage produces more data, which trains a better product, which attracts more usage.

A search engine with more searches learns which results satisfy people, so it returns better results, so more people search it. A recommendation system with more viewing history predicts better, so it recommends better, so people watch more. The loop runs through learning, not through users connecting to each other.

Data effects are real but frequently overstated, and the honest view matters. They have diminishing returns: the millionth example teaches a model far less than the thousandth, so past some point more data barely improves the product, and the advantage stops compounding. They are also often narrow, valuable only for the specific task the data describes. Watch-history data makes a better recommender and does nothing for an unrelated product.

So the test for a genuine data network effect is sharp: does more data keep making the product noticeably better, in a way a competitor cannot easily match by buying or collecting a smaller amount? Frequently the answer is no, and a claimed data moat is really just ordinary scale.

How fast does value grow with users? The famous answer is Metcalfe's law: the value of a network is proportional to the square of the number of users, because the number of possible connections between n people grows like n squared.

The intuition is right, value grows faster than user count, and the specific formula is almost certainly too optimistic, which matters because people use it to justify wild valuations.

The flaw is the assumption that every connection is equally valuable. It is not. You actively value a handful of close contacts, occasionally value some others, and place essentially zero value on the millions of strangers also on the network. Counting all n-squared possible pairs treats a connection to a stranger as worth the same as one to your closest friend.

More careful analyses, including work by Bob Briscoe, Andrew Odlyzko and Benjamin Tilly, argue growth is more like n times the logarithm of n, still faster than linear, but far below the square. The practical takeaway is not the exact exponent but the discipline: network value grows super-linearly, so scale matters enormously, but believing the n-squared version leads to overvaluing networks and underestimating how quickly the marginal user adds less. Real effects are strong and also saturate.

One distinction prevents most confusion in this area: network effects and switching costs both create lock-in, and they are different mechanisms with different implications.

A switching cost is a penalty for leaving: data trapped in a proprietary format, a contract, relearning a new tool, losing your history. It is a wall around the exit.

A network effect is a reason to stay that has nothing to do with leaving being hard: the product is genuinely more valuable to you because others are there. Even with a costless, instant exit you would not go, because the alternative, being on a smaller network, is simply worse.

The difference is strategic. Switching costs are a defensive moat that shrinks as technology makes moving easier and as regulators mandate portability. A network effect is a moat that deepens with success and that a competitor cannot dissolve by making migration easy, because the thing they lack is not a data-export button but the users themselves. That is why network effects are the more coveted and more durable advantage of the two.

Direct effects run within one group, indirect effects run between two groups, and data effects run through a learning loop. All three are flywheels; the last one tends to saturate.