Skip to content
Here Be Dragons

Network effects: Metcalfe's Law and Reed's Law

Two laws that explain why network-effects markets behave so differently from ordinary markets. Metcalfe’s Law describes how value grows with connections. Reed’s Law describes why group-forming accelerates that growth further. Together they explain why dominant platforms are so hard to displace, and why being second in a network-effects market is so costly.

Metcalfe’s Law

Origin: Robert Metcalfe, co-inventor of Ethernet, articulated this principle in the late 1970s. Formalized and named by George Gilder in 1993.

The value of a network is proportional to the square of the number of connected users.

The math: In a network of n users, each user can connect with n-1 others. The number of potential connections is n*(n-1)/2, which grows as O(n2)O(n²). Double the users, roughly quadruple the connections. That quadrupling of potential connections is what Metcalfe captured: value grows with connections, and connections grow quadratically.

Example 1: Fax machines

Metcalfe’s original illustration. One fax machine is worthless (no one to send to). Two fax machines have one connection. Ten fax machines have 45 connections. A million fax machines have roughly 500 billion potential connections. The value of owning a fax machine scales not just with your individual ability to send faxes, but with how many other people have fax machines. This is why fax adoption had a tipping point: past some critical mass, not having a fax machine became a competitive disadvantage.

Example 2: WhatsApp vs. feature-rich competitors

In 2012, WhatsApp had limited features: text messages, photos, and a contact list synced from your phone. Competitors had read receipts, group video calls, games, payment integrations. WhatsApp had more users. Users chose WhatsApp not because it was better but because their contacts were already on it. Network value dominated feature value. Facebook acquired WhatsApp for $19 billion in 2014, for the network, not the technology.

Example 3: Slack’s enterprise adoption

Slack’s growth within companies followed Metcalfe’s Law. A single team on Slack has low value. When the company adopts Slack, suddenly every team can reach every other team. The value did not grow proportionally to the number of Slack accounts. It grew much faster, because each new account added connections to all existing accounts. This is why “free tier to get teams on the platform” was Slack’s growth strategy: maximize n first, monetize the quadratic value second.

Example 4: Being second in a network-effects market

If platform A has 1,000 users and platform B has 100 users:

  • A’s connection value: roughly 500,000 pairs
  • B’s connection value: roughly 5,000 pairs

A has 100x B’s value even though A has only 10x the users. For B to compete, it needs to either provide 100x more features per connection (very difficult) or grow to near parity in users (very expensive). This is why network-effects markets tend toward winner-take-most. The math compounds against challengers.

Example 5: TCP/IP vs. competing protocols

In the 1980s, there were multiple competing networking protocols: OSI, IPX, AppleTalk, DECnet, TCP/IP. Each had technical merits. TCP/IP won not primarily on technical grounds but on network-effect grounds: the internet ran TCP/IP, the internet had the most nodes, and connecting to it required TCP/IP. The network value of TCP/IP was n² where n was the global internet. OSI’s n was comparatively tiny.

The limits of Metcalfe’s Law:

Not all connections are equal: Metcalfe’s Law counts potential connections. Real value depends on how many of those connections are actually useful. Twitter has 300 million users. For a given user, the valuable connections are maybe 500 accounts they follow. The other 299,999,500 connections exist but provide little direct value. Metcalfe’s original formula overestimates value in networks where most connections are irrelevant.

Spam and noise: As n grows, undesirable connections grow with it. Email was more valuable with 10 million users (mostly technical users who used it correctly) than it became at 4 billion users (because spam reached 85% of volume by 2020). Metcalfe’s Law counts connections. It doesn’t subtract noise.

Data and measurement: Metcalfe’s Law is difficult to verify empirically because “value” is hard to measure. Odlyzko and Tilly (2005) argued the actual growth is closer to n*log(n) for most networks, not n². The squaring assumes all potential connections are equally valuable, which isn’t true in practice.

Reed’s Law

Origin: David P. Reed, “The Law of the Pack,” Harvard Business Review, 2001.

The utility of large networks, particularly social networks, scales exponentially with the number of users (2^n).

Reed’s argument: Metcalfe’s Law counts pairwise connections. But in social networks, the dominant value comes from group-forming, not from 1-to-1 connections. Groups can include any subset of members. The number of possible subsets of n members is 2^n.

The example: A group chat app. With Metcalfe’s Law, value grows as n². But with Reed’s Law, the value comes from team chats, project groups, interest groups, regional groups — every possible subset. Those subsets grow as 2^n.

This explains why Facebook Groups, Slack channels, Reddit subreddits, and WhatsApp groups generate disproportionate engagement compared to purely 1-to-1 communication products. The group-forming utility dominates at scale.

The practical implication: If you’re building a platform, whether you support group formation changes your growth math. A messaging app (1:1 only) grows as n². A messaging app with group chats grows faster, potentially as 2^n for the group-forming value. This is why every major platform eventually adds group or community features.

How the two laws relate

Metcalfe describes the baseline. Reed describes the multiplier you get when groups are possible. In practice, 2^n is an upper bound; most possible subsets are never formed. But the directional claim holds: platforms that enable group formation consistently show higher engagement and stronger retention than those that only support 1-to-1 connections.

The practical question when designing a networked product: does your core interaction produce value for pairs (Metcalfe territory) or for groups (Reed territory)? The answer changes what features to build and what growth mechanics to optimize for.

References

  • Metcalfe, R. & Gilder, G. (1993). “Metcalfe’s Law and Legacy.” Forbes ASAP.
  • Reed, D.P. (2001). “The Law of the Pack.” Harvard Business Review.
  • Odlyzko, A. & Tilly, B. (2005). “A refutation of Metcalfe’s Law and a better estimate for the value of networks.” Minnesota Digital Technology Center Working Paper.
  • Shapiro, C. & Varian, H. (1999). Information Rules: A Strategic Guide to the Network Economy. Harvard Business School Press.
Why the name "Here Be Dragons"?