A network effect exists when each additional user makes the product more valuable to every other user. A network's value rises super-linearly with participants, while the cost of serving them rises roughly linearly. So past critical mass, the leader's product is better. Engineering has little to do with it. It's better because everyone else is already there. It's the only moat source that strengthens automatically with growth, so the widest modern moats are network businesses. That same pull makes it the most over-asserted claim in equity storytelling. There are three types:
- Direct (same-side): users value other users of the same kind. Telephones, WhatsApp, exchanges' liquidity (every trader wants the book where the other traders are).
- Indirect / two-sided: two distinct groups, each valuing the other's presence. Think buyers with sellers (marketplaces), merchants with cardholders (payments), developers with users (operating systems). The platform's power comes from refereeing the match and taking a toll, the take-rate machinery you've already studied.
- Data network effects: more usage produces more data, which makes a better product, which draws more usage. It's real in a few cases. Maps are one, and fraud scoring at Visa's scale is another. But it's the most overclaimed of all. Most "data moats" hit sharply diminishing returns, since the millionth training example improves the model less than the hundredth did. Data is often replicable or purchasable too. Demand extraordinary evidence.
Advertisement
What limits a network effect?
Where network effects cap out (the analyst's real job). Every network-effect claim has to survive four questions.
(1) Is the network global or local? Ride-hailing and food delivery clear at city level. A rider in Indore gains nothing from drivers in Chennai, so the "network" is 50 small networks, each contestable by a subsidizing entrant. It's why Ola/Uber and Zomato/Swiggy remain duopolies with negotiated-away margins instead of winner-take-all markets.
(2) Can users multi-home? If both apps sit on every phone and switching per transaction is one tap, the network effect sets a floor on relevance but not a ceiling on price.
(3) Does value actually rise with the marginal user, or does the network congest? More sellers can mean counterfeits and search costs. More riders at peak can mean surge pricing and worse service.
(4) Can regulation or public infrastructure neutralize the toll? UPI, below, is the case to study.
How strong is Visa's network moat?
Worked example (US/global): Visa. It's the strongest network moat in public markets. Its four-party model puts ~4.5bn credentials on one side and ~130m+ merchant acceptance points on the other. Each new cardholder makes acceptance more valuable, and each new merchant makes the card more valuable. An entrant offering a "better card network" with no acceptance is offering nothing. Try selling a card that no shop takes. That cold-start problem is the wall.
On illustrative ~FY24 figures (pull the 10-K), net revenue is ~$36bn on roughly $15 to 16tn of total volume, a toll of ~20 to 25 basis points. Operating margins run around 65%, and ROIC is far above any hurdle you could construct, stable for 15+ years. Instead of fragmenting, the market settled into a near-duopoly with Mastercard, which is what network economics predicts. Notice the stacking, too. Scale cost advantage reinforces the network effect, since the marginal transaction costs ~nothing. So does a brand that means "will work, everywhere, disputes resolved."
Can a government neutralise a network effect?
The UPI counter-case (India, drill this). Now watch a government neutralize that very moat. India's UPI is public rails built on the India Stack, interoperable by design. By policy since 2020 it carries zero MDR (merchant discount rate) on P2M payments. It now moves on the order of ₹260 lakh crore a year, roughly $3tn at FY25 scale. Verify that at NPCI's statistics page, which publishes monthly. UPI reproduced the two-sided network, because every QR code accepts every app and every app reaches every bank account.
It just did so as public infrastructure with the toll set to zero.
Look at what followed for analysts. India's domestic debit-card economics were largely erased. Card networks' Indian growth story narrowed to credit, where MDR survives, including RuPay credit-on-UPI. Fintechs riding UPI (Paytm, PhonePe, Google Pay) carry enormous volume with no toll.
A network without a tollbooth is traffic, and traffic alone isn't a business.
Three lessons from this are permanent. (1) A network effect is a mechanism and not a birthright. Its monetization can be regulated away even while the network itself thrives. (2) "We have millions of users transacting" is worthless without a defensible take rate. Keep GMV and net revenue apart, as you learned to with revenue models. (3) In India specifically, always ask whether the state could decide your toll is public infrastructure. In fairness, policy debates about reintroducing MDR for large merchants recur. So the method is to check current NPCI/MeitY policy. Don't memorize today's.
Is IndiaMART a real network moat?
Worked example (India): IndiaMART. Here's a domestic two-sided network at profitable scale. IndiaMART holds ~60%+ share of B2B classifieds, matching tens of millions of buyer enquiries to millions of supplier storefronts. Buyers come because the suppliers are there. Suppliers pay subscriptions because the buyers are there. Illustratively, at ~FY25 scale (pull the annual report), revenue is ~₹1,350 to 1,400 crore with EBITDA margins of ~35%+. Customers pay in advance, so deferred revenue creates negative working capital, and invested capital in the operating business is near zero. That makes ex-cash ROCE extreme. The past spread isn't in doubt. The moat question is the cap-out checks: supplier churn in the long tail, and whether high-intent categories peel off to vertical rivals.
It's a real network moat, sized narrow, and sized fairly.
How do you test a network effect?
Evidence trail for network effects (where to look). Track share of transactions or volume over time alongside take rate. Share up with take rate up means network power. Share up with take rate down means the company is buying traffic. Next, read liquidity metrics by cohort (match rates, time-to-fill, enquiries per listing). Check multi-homing data such as app-install overlap and merchant acceptance of rivals. Look at cohort retention on both sides, and at whether contribution margin per transaction trends up with scale. Then run the regulatory scan. Is the toll itself a policy variable?