Network Effects: The Most Powerful Moat Explained
A network effect exists when every new user makes a product more valuable to every existing user. It is often called the most powerful moat because, unlike a patent or a cost edge, it can get stronger as the business grows, pushing markets towards one or two dominant players.
Network effects sit behind some of the most profitable businesses ever built, from card networks to stock exchanges. They are also the most overused phrase in investing. Plenty of companies with millions of users have no network effect at all, and learning to tell the difference will save you from paying a premium for a moat that is not there.
Where this fits: this is a deep dive on one of Pat Dorsey's five moat sources for Criterion 2, "Does it have a durable moat?" Start with our moat hub for the full picture, and see how moats fit the whole process in the Five Criteria framework.
What Is a Network Effect?
A network effect is a feedback loop between users and value. More users make the product more useful. A more useful product attracts more users. Once that loop is spinning, a rival has to offer something dramatically better just to get a hearing.
The classic illustration is the telephone. One phone is useless. Two phones allow one conversation. Every additional phone adds a new connection for everyone already on the network. Ethernet inventor Robert Metcalfe later popularised the idea that a network's value grows much faster than its user count, an idea now known as Metcalfe's law. Treat it as a rough intuition, not a formula you can value a company with.
Pat Dorsey, in The Little Book That Builds Wealth, points out that network effects are most common in businesses built on sharing information or connecting people, rather than in businesses that make physical goods. That is a helpful first filter.
Why network effects are so powerful
Most moats are static. A brand must be defended with advertising. A patent expires. A cost advantage can be copied by a rival who builds the same factory.
A network effect can compound instead. Hamilton Helmer, in 7 Powers, makes this point: the leader's advantage can widen as it grows, because each new user adds more value to the larger network than to the smaller one. Markets with strong network effects therefore tend to tip towards one or two winners.
The Main Types of Network Effects
Not every network effect is the same strength. Knowing the type tells you how durable the moat is likely to be.
1. Direct (same-side) network effects
Users benefit directly from other users of the same kind. Messaging apps and social networks are the standard examples. You join because your friends are there, and your joining makes it more valuable for them.
Direct effects can be very strong, but they can also unwind quickly if users move as a group. MySpace was the leading social network in the mid-2000s and lost that position to Facebook within a few years. A direct network is only as sticky as the relationships inside it.
2. Indirect (two-sided or cross-side) network effects
Two different groups benefit from each other. More buyers attract more sellers, and more sellers attract more buyers. Card networks, marketplaces, app stores and video game consoles all work this way.
Two-sided networks are often more durable than direct ones. To break in, a rival must win both sides at once, which is a classic chicken-and-egg problem. A merchant will not accept a new card that no one carries, and consumers will not carry a card that no merchant accepts.
3. Liquidity network effects
Financial exchanges have a special version. Buyers and sellers go where the most other buyers and sellers already trade, because that is where prices are tightest and orders fill fastest. Liquidity attracts liquidity.
This is why listing venues and derivatives exchanges can hold dominant positions for decades. A trader rarely moves volume to a new venue with thinner prices just to save a small fee.
4. Platform and ecosystem effects
Some networks are built around developers or complementary products. The more users a platform has, the more developers build for it. The more software exists for it, the more attractive it is to users. Microsoft Windows on the personal computer is the best-known historical example.
5. Data network effects (handle with care)
Some companies argue that more users generate more data, which improves the product, which attracts more users. Sometimes this is real. Often the benefit flattens quickly, because the hundred-millionth data point adds little that the ten-millionth did not. Be sceptical when a company leans on "data network effects" without evidence that the product actually improves.
Local vs global networks
A final distinction matters a great deal. Some networks are global: a card works almost everywhere. Others are local: a ride-hailing driver in Toronto is of no use to a rider in Montreal. Local networks can be attacked city by city, so they tend to be weaker moats than their user numbers suggest.
Type | Who benefits | Typical durability |
Direct | Same group of users | High, but can unwind in a group move |
Two-sided | Buyers and sellers | Very high when both sides are large |
Liquidity | Traders and issuers | Very high |
Platform | Users and developers | High while the platform stays relevant |
Data | Users, via product quality | Often weaker than claimed |
Real-World Examples of Network Effects
The examples below are described in terms of their structure only. Always check the latest annual report before drawing conclusions about any company's current numbers.
Visa and Mastercard
The card networks are the textbook two-sided network. Visa began life as BankAmericard, launched by Bank of America in 1958, and took the Visa name in 1976. Its value to a consumer depends on how many merchants accept it, and its value to a merchant depends on how many consumers carry it.
Importantly, Visa and Mastercard do not lend money. Banks issue the cards and carry the credit risk, while the networks earn fees for processing transactions. That makes them capital-light, which is exactly what a strong network moat should look like in the numbers.
Stock exchanges
Exchanges such as those run by Intercontinental Exchange, Nasdaq, CME Group and the London Stock Exchange Group benefit from liquidity network effects. Companies want to list where investors trade, and investors trade where companies list. Derivatives contracts are particularly sticky, because open positions and trading habits concentrate in a single contract.
Online marketplaces
eBay, founded in 1995, showed how a marketplace network works: collectors listed items where the most buyers looked, and buyers looked where the most items were listed. It also shows the limits. Amazon's decision to open its site to third-party sellers built a second large marketplace alongside eBay, proving that a network effect in one category does not protect a company in every category.
Microsoft Windows
For decades, the huge base of Windows users encouraged software developers to write for Windows first. That library of software then kept users on Windows. This platform effect is a large reason the operating system held its position on desktop computers for so long.
A cautionary tale: MySpace
MySpace was once the most visited social network in the United States. News Corporation bought it in 2005. Within a few years users had moved to Facebook, and News Corp sold MySpace in 2011 for a small fraction of what it paid. A large network is not a moat on its own if users can move together to something they prefer.
How to Test a Network Effect with Numbers
A network story is a hypothesis. These tests turn it into evidence.
1. Returns on capital that rise with scale
A genuine network effect should show up as high and rising returns on invested capital as the business grows. The house test from Criterion 1 still applies: ROIC of 15% or more for five years or longer. Our ROIC guide walks through the calculation.
2. Operating margins that expand
Networks usually have high fixed costs and low costs per additional transaction. If the network is working, operating margin should widen as volume grows. Flat margins despite rapid growth suggest the company is paying for every new user.
3. Stable or rising take rates
The take rate is the share of each transaction the network keeps. A strong network can hold or raise its take rate without losing volume. A falling take rate, especially while volume grows, often means users have alternatives.
4. Falling customer acquisition costs
In a working network, word of mouth and the value of the network itself do much of the selling. Watch sales and marketing expense as a share of revenue. If it rises year after year just to keep growth steady, the network effect is weaker than the story.
5. Market share that concentrates
Network markets tend to tip. Is the leader gaining share over time, or are several players holding steady? Stable fragmentation suggests either weak network effects or easy multi-homing.
6. Gross margin through a downturn
The downturn test from the moat hub is still the most important one. A true network should hold its gross margin and pricing when the economy weakens, because users have nowhere better to go.
Worked Example: Two Marketplaces
Here are two hypothetical online marketplaces, each reporting five years of results. Both claim network effects.
Year | Marketplace A take rate | Marketplace A ROIC | Marketplace B take rate | Marketplace B ROIC |
Year 1 | 10% | 14% | 12% | 11% |
Year 2 | 10% | 18% | 11% | 12% |
Year 3 (recession) | 10% | 17% | 9% | 5% |
Year 4 | 11% | 21% | 9% | 7% |
Year 5 | 11% | 24% | 8% | 8% |
Marketplace A kept its take rate steady through the recession and nudged it up afterwards. ROIC rose as volume grew. That is what a network effect looks like in the data.
Marketplace B started with a higher take rate, but it had to keep cutting it to hold volume. Its returns fell sharply in the recession and never recovered. Sellers were clearly listing on other sites too, and B had to compete on price. Whatever it says in its investor presentation, B does not have a working network effect.
How Network Effects Erode
Network moats are powerful, but they are not permanent. These are the main ways they weaken.
Multi-homing. If users can easily use two networks at once, the network effect weakens. Drivers and riders who run two ride-hailing apps side by side are a common example. The more multi-homing, the less pricing power.
Disintermediation. Once two sides have met through a platform, they may deal directly next time. This is a constant risk for marketplaces connecting repeat buyers and sellers, such as freelance services.
Congestion and quality decline. A network can become too crowded, filled with spam, poor listings or low-quality content. Past a point, more users make the product worse, not better.
Regulation. Dominant networks attract regulators. The European Union, for example, capped consumer card interchange fees in 2015 at 0.2% of the transaction for debit cards and 0.3% for credit cards. Rules like these can cap what a network earns even if its position is intact.
Platform shifts. A new technology can create a new network that makes the old one less relevant. The move from desktop computers to smartphones reduced the importance of the desktop operating system for many everyday tasks.
The network gets out-grown. If the network serves a niche that stops growing, the effect protects a shrinking pool of profit.
Having worked in investor relations, I can tell you that "network effects" is one of the phrases management teams love to include in a strategy slide. Ask for the take rate, the share trend and the marketing spend. Those numbers rarely lie.
How We Use This in the Five Criteria
Criterion 2 asks: what stops competitors eroding this company's returns? A company passes our house default when you can name one of Dorsey's five sources and show evidence that ROIC held through a downturn.
For a network effect, that means:
Name the type: direct, two-sided, liquidity, platform or data, and say whether it is local or global.
Explain it in two sentences: who benefits from whom, and why a rival cannot easily recreate it.
Show the evidence: ROIC at or near 15% through the industry's worst recent year, with a stable take rate and gross margin.
A network business that passes Criteria 1 and 2 still needs to pass the rest. Many famous network companies trade at high valuations, so Criterion 5 and the margin of safety often become the binding constraint.
Criterion 2 checklist for network effects
Can I name the type of network effect and the two sides (if any) in one sentence?
Does the product clearly become more useful to existing users as new users join?
Is the network global, or can a rival attack it one city or niche at a time?
Is the leader gaining market share over time?
Has the take rate been stable or rising for five years or more?
Is sales and marketing spend flat or falling as a share of revenue?
Did ROIC stay at or near 15% in the industry's worst recent year?
How easy is it for users to multi-home or go direct?
Is there a regulator or a new technology that could cap or bypass the network?
If you answer yes to the first seven and are comfortable with the last two, you have strong evidence of a network moat.
Common Mistakes When Analysing Network Effects
Confusing scale with a network effect. A streaming service with 100 million subscribers does not become more useful to me because you subscribe. That is scale, and it belongs under cost advantages, not networks.
Ignoring multi-homing. If users happily use several competing networks, the moat is far weaker than the user count suggests.
Assuming the leader always wins. MySpace shows that direct networks can unwind when users move together.
Counting users instead of transactions. Registered users are easy to inflate. Active users, transaction volume and revenue per user tell you more.
Paying any price for the story. The best network businesses are well known and rarely cheap. A great moat bought at a silly price can still be a poor investment.
Believing "data network effects" without proof. Ask how much the product actually improves with each additional user.
Frequently Asked Questions
What is a network effect in simple terms? It is when a product becomes more valuable to each user as more people use it. Card networks, marketplaces and exchanges are classic examples, because each new participant makes the network more useful to everyone already on it.
Why are network effects considered the strongest moat? Because they can widen as the business grows, rather than weaken over time. Markets with strong network effects tend to tip towards one or two winners, and a new entrant must win both sides of the market at once to compete.
What is the difference between direct and indirect network effects? Direct effects happen within one group, such as friends on a messaging app. Indirect, or two-sided, effects happen between two groups, such as buyers and sellers on a marketplace, who each benefit from the other side's growth.
Can a network effect moat be broken? Yes. Multi-homing, disintermediation, congestion, regulation and new technology platforms can all weaken a network. Watch the take rate, market share and marketing spend for early signs.
Your Next Step
Take one company you believe has a network effect and run it through the nine-point checklist above. Then compare it with the other four moat sources in our moat hub, and read how Nick Sleep thought about compounding advantages in Nick Sleep and Nomad: scale economies shared. When you are ready, score the full business with the Five Criteria Checklist.
About the author: Cameron Hayes is a senior investor relations and finance professional who has worked in IR at Nasdaq Copenhagen-listed companies including Nilfisk and Maersk Drilling. He holds an MSc from Copenhagen Business School and a BCom from the University of Ottawa, and founded Gingernomics to teach individual investors the Five Criteria framework. More about Gingernomics.
The content on Gingernomics is for educational and informational purposes only and does not constitute financial advice. Always do your own research and consult a licensed financial advisor before making any investment decisions. Past performance is not indicative of future results.
Comments