Network Effects: The Only Moat That Deepens While You Sleep

A fax machine sitting alone in an office in 1975 was an expensive paperweight. It could do nothing at all. The second fax machine made both slightly useful. By the time there were a hundred thousand, refusing to own one had become a competitive disadvantage — and the machines themselves had not changed in any way.

This is the strangest property in business economics. Almost every product improves because someone improved it: engineers ship features, manufacturers refine tolerances, designers iterate. A network-effect product improves because more people showed up. No one built anything. The value arrived as a by-product of adoption.

Understanding this properly explains why certain companies become nearly impossible to dislodge despite building products that are, feature for feature, unremarkable — and why far more companies claim to have network effects than actually possess them.

The mechanism, and what it is not

The precise claim is: the value of the product to each user increases as more users join. That "to each user" clause is doing critical work, and it is where most loose usage of the term collapses.

A popular restaurant with a queue is not a network effect. It has scale, brand, and perhaps supply advantages — but your dinner is not improved by other diners; if anything it is degraded by the wait. A product with many customers is not thereby a network. The test is causal and specific: does an additional user make the product better for the users already there?

Where the answer is genuinely yes, the consequences compound in a way scale alone never does:

Growth becomes self-reinforcing. More users make the product more valuable, which attracts more users. This is a positive feedback loop, with all the runaway behaviour that implies.

Competition inverts. In a normal market, a superior product wins customers. In a network market, a superior product with fewer users can be genuinely worse for any individual to use. A chat application with better design and none of your contacts is not a better chat application — it is a worse one, for you, today. This is why network businesses are so resistant to being out-engineered.

The moat strengthens with size. Physical moats erode: patents expire, factories age, brands fade. A network moat does the opposite — every additional user deepens it, at no cost to the owner.

Four types, unequally defensible

Lumping all network effects together is the second common error. They differ enormously in how attackable they are.

Direct (same-side) effects are strongest. Every user makes the product better for every other user — telephones, messaging, social graphs. Dislodging one requires convincing not just a person but their entire circle to move simultaneously.

Two-sided (marketplace) effects connect distinct groups: buyers and sellers, riders and drivers. Powerful, but structurally weaker, because a competitor can attack one side at a time, and because both sides frequently multi-home — a driver running two apps, a seller listing on three marketplaces. Multi-homing is the single most reliable solvent for marketplace network effects.

Local effects are geographically bounded. A ride-hailing network is only valuable to you in your city, which means a competitor need only win one city, not the world. Businesses that look globally dominant on this basis are often a portfolio of separately contestable local monopolies.

Data effects — the product improves as usage generates data that improves the algorithm — are the most over-claimed. They are real but usually saturate: the thousandth data point improves the model far more than the billionth, so the advantage plateaus well before it becomes insurmountable.

Metcalfe's Law formalised the intuition that value scales with the square of users, since possible connections do. The literal formula overstates reality — most users never interact with most others, and later users are typically less valuable than earlier ones. But the qualitative claim survives: value grows faster than linearly with users, and that superlinearity is the whole game.

The cold start problem, and how networks actually die

Every network effect contains a brutal paradox. The property that makes the product unbeatable at scale makes it worthless at launch: the first users receive almost no value, because there is no network yet. Nearly every network business's real origin story is not about the network — it is about what got the first users to stay while it did not yet exist.

The historically effective answers are consistent. Deliver standalone value first — be useful to a single user with no network, and let the network become a bonus that accrues later. Constrain the initial market ruthlessly — a network that is dense within one campus, one city, or one profession is genuinely valuable, whereas the same number of users scattered globally is worth nothing. Density beats size at the start. Subsidise the harder side — in marketplaces, one side is usually scarcer, and paying to acquire it is often the entire early strategy.

And networks do die, which cuts against the mythology of permanent dominance. Three things kill them:

Platform shifts. A network built for one computing environment must be rebuilt for the next. The transition is the vulnerable moment, because habits are re-formed and the incumbent's advantage briefly resets — the same window that makes the innovator's dilemma so dangerous.

Negative network effects. Growth is not monotonically good. Past a threshold, more users can mean more noise, more spam, more congestion. Any network whose value depends on the quality of interactions can be degraded by its own growth.

Segment defection. Networks rarely collapse all at once; they lose a coherent sub-group that valued something the mainstream network stopped providing. That sub-group forms a dense small network elsewhere, and the process repeats.

The practical test, when someone claims a network effect, is to ask one question: if this product had ten users instead of ten million, would it be worse — not less popular, but functionally worse? If yes, the moat is real and it deepens daily. If the honest answer is "it would be equally good, just less well known," you are looking at a brand, and brands can be out-spent.