Metcalfe's Law: Why Second Place Is So Much Worse Than Second

Robert Metcalfe, who worked on early networking technology, proposed a rough rule for why networks become valuable: the value of a network grows with the square of the number of users.

The reasoning is simple counting. In a network of n members, the number of possible pairwise connections is roughly n²/2. Ten users produce forty-five possible pairs. A hundred users produce nearly five thousand. A thousand produce almost half a million.

So if the value of a network comes from connections between members, then adding members increases value faster than it increases the member count — ten times the users gives roughly a hundred times the possible connections.

This is the mathematical statement of why network effects are so powerful, and the exact formula is wrong in ways worth understanding.

What the formula gets right

Before the criticism, the core claim deserves credit, because it captures something genuinely important that simpler models miss.

Value is superlinear in users. Whether it grows as the square or by some gentler curve, network value grows faster than proportionally. This is unlike almost every other kind of business, where doubling customers roughly doubles the value delivered.

Small networks are worth very little. The formula produces near-zero value at small sizes, which matches reality — a network with ten members is usually not one per cent as useful as one with a thousand. This is the arithmetic behind the cold-start problem.

Merging two networks creates value from nothing. Combining two equal networks more than doubles the connections, because members of each can now reach members of the other. This explains why interoperability and standards matter so much, and why controlling whether networks can interconnect is a significant form of power.

Being second is much worse than being slightly behind. If value scales with the square, a competitor with half the users has roughly a quarter of the value, not half. Small leads compound into large gaps.

Why the exact formula overstates

The square relationship assumes every possible connection is equally valuable, and that is clearly false.

Nobody interacts with most members of a large network. On any large platform, the typical person has meaningful contact with a small number of others — which is the constraint described in Dunbar's number. The remaining millions of theoretically possible connections contribute essentially nothing to that person's experience.

Connections also vary enormously in worth. Reaching one specific person may be worth more than reaching ten thousand strangers.

And later users are usually less valuable than early ones. The people who join first tend to be the most engaged and best connected; growth eventually reaches people who use the network rarely.

Alternative formulations have been proposed suggesting value grows more slowly than the square while still growing faster than linearly. The empirical work is mixed and depends heavily on how value is measured.

The sensible reading is that the exact exponent is unresolved and the qualitative claim is robust: network value grows faster than user count, small networks are disproportionately weak, and leads compound. Treating the specific formula as a valuation tool has produced some notably bad forecasting, particularly during periods when it was used to justify prices for networks that had users and no revenue.

Where the reasoning does not apply

The formula assumes a network where any member can usefully reach any other, and where more members improve things. Both fail in identifiable cases.

Broadcast is not a network in this sense. If information flows one way from a centre to many recipients, value scales roughly linearly with audience, because recipients do not connect to each other.

Local networks do not aggregate globally. Where the useful connections are geographically bounded — riders and drivers in a city — the relevant network is the city, not the whole system. A business with users in fifty cities is closer to fifty networks than one large one, which is a much weaker position than the global user count suggests.

Growth can add noise. Where value depends on the quality of interactions rather than their possibility, more members can make a network worse. Communities frequently degrade past a certain size, and this is not a failure of moderation so much as a predictable consequence of scale changing the composition of participants.

Multi-homing dilutes the effect. If members participate in several competing networks simultaneously, no single network captures the value its user count implies.

Held with those caveats, the underlying insight remains one of the more useful in business strategy: in a network, the arithmetic works against second place, and that is why so much competitive behaviour in these markets looks disproportionate to what is immediately at stake.