Network Effects and Switching Costs: The Modern Economy's Favourite Moats

Between them, these two mechanisms explain more of the modern economy's concentration than any other entry in the moat catalogue — why a handful of platforms dominate social media, marketplaces, operating systems, and payments; why software is sold as subscriptions wrapped in ecosystems; and why challengers with better products routinely fail to displace incumbents with bigger networks. This article separates the varieties (they behave differently), explains the winner-take-much dynamics and the counterforces that limit them, and closes with the honest section the hype omits: how these moats erode and reverse. Frameworks throughout, never verdicts on any firm.

Network effects: three varieties worth separating

Direct network effects: users value other users directly — telephones, messaging apps, social networks. Each new user makes the product better for every existing one, a compounding loop the early telephone industry already understood. Indirect (two-sided) network effects: two different groups value each other — riders and drivers, buyers and sellers, developers and device owners, merchants and cardholders. Platforms and marketplaces live here, and the Karsavia taxi example's third act ran on exactly this: each side joins the network where the other side already is, so leadership on one side recruits the other in a spiral. Category examples — ride-hailing, e-commerce marketplaces, app stores, payment networks — are factual illustrations, not endorsements. Data network effects, claimed widely in the AI era, deserve the most scepticism of the three: usage generates data that improves the product, which attracts usage — a real loop in some businesses, but one that often exhibits diminishing returns (the millionth data point teaches less than the thousandth), making many claimed data moats shallower than pitch decks suggest — a live analytical debate flagged as such. The economics underneath all three: network effects put the elasticity determinants on a conveyor — every added user deepens the no-good-substitute condition for everyone else.

Switching costs: the four flavours of staying put

Switching costs work on the individual customer rather than the crowd. Financial: exit fees, lost loyalty balances, repurchasing what you already own. Procedural: the time and risk of migration — retraining staff, rewriting integrations, re-certifying vendors; in enterprise software this alone sustains decades-long relationships, since the migration's risk lands on whoever approves it (an incentive structure favouring incumbency). Relational: accumulated trust, service history, and the human relationships around a product. Data and ecosystem: the modern heavyweight — files, histories, playlists, reviews, reputations, and integrations that don't port; every additional product in an ecosystem multiplies the cost of leaving all of it. Strategy decodes accordingly: subscriptions, bundles, proprietary formats, and loyalty programmes are switching-cost engineering (the elasticity article's "make the demand curve steeper" question, answered mechanically), and regulators' recurring interest in data portability and interoperability is the same mechanics read from the consumer's side — a policy debate this pillar's closing article treats two-sidedly.

Winner-take-much — and the honest limits

Where network effects run strongly, competition is for the market rather than in it: the prize is becoming the standard, which is why platform contests feature years of subsidised losses (buying the network the zero-marginal-cost economics then monetise) followed, for winners, by concentration and margin. But "winner-take-all" overstates a real pattern, and three counterforces deserve equal billing. Multi-homing: where using two platforms is cheap — drivers running two apps, sellers listing on two marketplaces, viewers holding three streaming subscriptions — no single network locks the market, and leadership stays contestable. Niche differentiation: networks fragment along communities, geographies, and use cases; the biggest network is not the best network for every group, so ecosystems of specialised platforms persist beside giants. Reversal: network effects run both ways — a shrinking network sheds value with the same compounding it once gained, which is how dominant social platforms, messaging services, and marketplaces of earlier eras emptied with startling speed once departures began; congestion, spam, and declining quality can start the spiral from inside. The analytical summary: a network moat is real where joining is valuable, leaving is costly, and multi-homing is impractical — three conditions to check separately, none permanent, with the durability question belonging to the disruption article.