The Network Effect Spectrum: Not All Are Created Equal

NFX's research across 336 companies found that 70% of all value created in technology since 1994 was generated by businesses with network effects. The problem is that most founders using the term have never stress-tested whether they actually have one.

Network effects are the most overused phrase in venture. They are also, when real, the most powerful economic force in business. The gap between claiming one and building one is where most startups fail — and most investor decks mislead.

The Four Types — And Where They Fall on the Spectrum

What Most Startups Are Actually Describing

When founders say "we have network effects," they usually mean one of these three things — none of which are network effects:

• ✕

  Switching Costs

  Your product is painful to leave because of data lock-in, workflow integration, or migration complexity. High-quality switching costs are real and valuable — Salesforce has them — but they don't make the product better with more users. They just make leaving painful.

• ✕

  Viral Loops

  Your product spreads because users invite others. Dropbox's referral program was brilliant growth — but the product wasn't more valuable because more people used it. Virality drives acquisition; it does not create defensibility.

• ✕

  Economies of Scale

  You get cheaper to operate as you grow, so you can undercut competitors on price. AWS has this. But price leadership from scale is different from a network — a better-capitalized entrant can always subsidize adoption to neutralize it.

Why Data Network Effects Are the 2026 Moat

Before large language models, the strongest network effect was direct — Facebook had 3 billion users who would never leave because everyone they knew was already there. Post-ChatGPT, the landscape shifts.

The emerging moat in AI-native companies is proprietary behavioral data that cannot be scraped, replicated, or synthesized. Consider:

Each of these companies has a moat that gets harder to breach every quarter — not because their UI is beautiful or their support is good, but because their model improves with usage in ways a new entrant cannot replicate on day one.

The Honest Test: Does Adding One More User Make the Product Better?

This is the only question that matters. Not "does adding more users grow revenue?" Not "does adding more users reduce churn?" Does the product itself get meaningfully better for every existing user when a new one joins?

• Scenario: LinkedIn adds a new senior developer in your city

  Your recruiter search results improve → direct network effect ✓

• Scenario: Waze adds 100,000 new commuters in Dallas

  Route accuracy improves for all Dallas users → data network effect ✓

• Scenario: Salesforce adds a new customer in your industry

  Your CRM doesn't change at all → switching cost, not network effect ✗

• Scenario: Dropbox adds a new user in your company

  Your file storage is unaffected → viral loop, not network effect ✗

• Scenario: Uber adds a new driver in your city at 9pm

  Your wait time drops → indirect network effect ✓

Network Effects and AI: The New Risk

Here is the uncomfortable truth for founders who think their marketplace network effect is permanent: AI lowers the cost of supply-side acquisition to near zero in several categories.

A legal marketplace that took 10 years to recruit 50,000 attorneys can be disrupted by an AI-native platform that automates 80% of the work those attorneys did — not by competing for the same supply, but by eliminating the need for it. The indirect network effect (attorney supply ↔ client demand) evaporates when the supply side becomes synthetic.

This is why I back companies where the network effect is on the demand side or in the data layer — not in aggregated human supply that can be automated away.