# What Counts as a Network Effect A true network effect makes the leader's cost to add the next user fall *relative to a smaller competitor*. A self-reinforcing loop alone does not qualify. — Martin Casado^[[x.com/martin_casado/status/2111870054891749501](https://x.com/martin_casado/status/2111870054891749501), 18 Oct 2026] ## The test: leader vs. challenger - **Network effect**: value grows faster than the user count. The leader's advantage widens as it grows. Example: telcos. - **Self-reinforcing property**: the company gets something per customer, such as data. Value grows **sublinearly**. Each extra unit is worth less to the leader than to a challenger, so the challenger closes the gap. - Casado's verdict: on such properties, the leader is "strictly disadvantaged" relative to a challenger. > [!question] Nuance > - "Strictly disadvantaged" refers to the marginal return. In absolute terms the leader still has more. What shrinks is the gap. > - Casado says network value grows "exponentially". [[Metcalfe's Law]] is quadratic (n²). Only [[Reed's Law]] is exponential (2ⁿ). ## No working consensus Casado's follow-up lists the definitions in use: 1. A derivative of [[Metcalfe's Law]] 2. A scale effect 3. Product value grows with the number of users 4. Definition 3 plus a flywheel 5. A flywheel effect 6. Value comes from an actual interconnected network (telco, ISP, social network, GitHub). The interconnections define the value. - He holds definition 6. Definitions 2, 4 and 5 describe other things. - His closing line: words mean whatever you want "if you don't care about communicating." ## Counter-frame: Tren Griffin Griffin uses the broader definition 3: value depends on the number of users. He adds two axes: | Axis | Kinds | Example | |---|---|---| | Sign | Positive / negative | Telephone network / congestion | | Path | Direct / indirect | Telephone / complements such as phone cases | - A network effect can protect a market of little financial value. Strength is not size. - His chart plots the four growth curves: Sarnoff (n), n log n, Metcalfe (n²), Reed (2ⁿ). See [[Network Laws]]. ## Use it in diligence 1. Does the leader's cost to add a user fall vs. a smaller rival? 2. Does value come from links *between* users, or from data *about* users? 3. Is the curve super-linear or sublinear? 4. Is it positive, or does congestion turn it negative? 5. Is it direct, or does it run through complements? If the answer to 2 is "data about users", call it a [[Data Flywheel]] or [[Data Moat]], not a network effect. > [!warning] Tension in this vault > [[Defensibility Principles MOC]] gives "ChatGPT getting better with more user queries" as a network effect. Under Casado's test, that is a data loop. ## Coverage - Read: the main post, Casado's two follow-ups, Tren Griffin's reply (2026-10-19). - Not read: ~55 further replies, which need an X login. ---- tags: #boat #defensibility links: [[Defensibility Principles MOC]] [[7 Powers]] [[Network Laws]] [[Metcalfe's Law]] [[Data Flywheel]] [[Data Moat]] [[AI era Defensibility]]