Types of moats: the 10 moats a startup can build

1. 01Network Effects: What They Are, the Main Types, and Real ExamplesWhat a network effect is, the main types (direct, marketplace, data, platform), 13 real examples, and how startups build one from zero.
2. 02Atomic networks: win a small market before you try to win a big oneWhat an atomic network is, why the cold start problem kills new networks, and how Facebook, Slack, Tinder, Uber and DoorDash won a small market first.
3. 03Owned Customer Relationships: Why Leakage Kills StartupsWhy startups that own the customer relationship keep their margin, and how marketplaces lose it to leakage. Examples: Airbnb, Upwork, Homejoy, Shopify.
4. 04Copycat competitors: how to protect a startup from being copiedWhy copycat competitors win or lose, with 10 real cases from Instagram to Figma, and how to build a startup a $50M rival still could not beat.
5. 05Proprietary Data Moats: When Data Actually Protects a StartupWhat a proprietary data moat is, when data moats are real, 12 company examples, the a16z skeptic view, and how startups collect data rivals can't buy.
6. 06The Data Flywheel: When Usage Makes the Product Measurably BetterWhat a data flywheel is, how it differs from data network effects and Amazon's flywheel, 11 real examples, and why more data often stops helping.
7. 07Switching Costs: Why Customers Stay, and How to Earn ItWhat switching costs are, the 7 types, 15 real examples from Salesforce to Epic and Figma, how startups build them, and how to tell real lock-in from fake.
8. 08Distribution Moat: How Startups Own Channels Rivals Can't BuyWhat a distribution moat is, owned vs rented channels, 12 real examples from HubSpot to PayPal, and how startups build distribution rivals can't buy.
9. 09Platform Risk: What Happens When the Platform Ships Your FeaturePlatform risk is the chance that Apple, Google, OpenAI, Meta or Shopify ships your product as a feature. See 14 real cases and how startups reduce it.
10. 10AI Moats: What Survives When Everyone Has the Same ModelWhat an AI moat is, why GPT wrappers get copied, and the empty database test: would your product survive a rival using the same model with no data?

A moat is an advantage that protects a business after competitors copy its product. There are many types of moats, but for a startup ten matter most: network effects, a winnable starting market, owned customer relationships, competitive resilience, proprietary data, a data flywheel, switching costs, owned distribution, platform independence and value beyond the AI model.

The term comes from investing. Warren Buffett wrote in his 2007 letter to Berkshire Hathaway shareholders that "a truly great business must have an enduring 'moat' that protects excellent returns on invested capital," and that "a moat that must be continuously rebuilt will eventually be no moat at all." Hamilton Helmer's 7 Powers gives a more formal list for strategists: scale economies, network economies, counter-positioning, switching costs, brand, cornered resource and process power.

Those frameworks judge established companies. A startup's question is sharper: once the idea works, what stops someone bigger from copying it? The ten moats below can each be checked at the idea stage. For the basics, start with what a moat is.

The 10 types of moats

1. Network effects

The product gets better for each user as more people use it. Marketplaces, messaging and payment networks work this way, and a rival has to recreate the whole network, not just the product. NFX estimates that network effects have been responsible for 70% of the value created in technology since 1994. Read the full guide to network effects.

2. A winnable starting market (atomic network)

Durable networks start small: one team, one campus, one city or one niche, dominated before expanding. Without a small market that works on its own, a network product cannot get past the cold start. See atomic networks.

3. Owned customer relationships

If customers meet in your product and then move to email, Slack or someone else's marketplace, that platform holds the leverage. The moat forms when interactions, payments and records live inside your product. See owned customer relationships.

4. Competitive resilience

The blunt test: could a rival with $50M and a better AI model, launching in 12 months, still fail to win the same users? If not, the idea is a feature waiting to be copied. See copycat resistance.

5. Proprietary data

Normal use creates data that competitors cannot buy, scrape or synthesize: outcomes, transactions, local or perishable information. Public or purchasable data is not a moat. See proprietary data moats.

6. Data flywheel

Data only becomes a moat when it measurably improves the product as usage grows. Even then, returns often diminish, as a16z argues in The Empty Promise of Data Moats. See data flywheels.

7. Switching costs

Leaving would mean losing history, integrations, workflows or reputation. CRM, payroll, accounting and design systems build this naturally, because years of work live inside them. See switching costs.

8. Distribution advantage

A channel to customers that rivals cannot simply buy: a community, an audience, a partnership or an integration inside the customer's daily workflow. Paid ads and app stores are rented distribution, available to anyone with a budget. See distribution moats.

9. Platform independence

If Apple, Google, OpenAI, Meta or Shopify could ship your product as a feature, you have no moat against them. Building on a platform is often necessary; depending on it for your core value is the risk. See platform risk.

10. Value beyond the AI model

If a competitor could plug the same model into an empty database and match you, the value belongs to the model provider, not your business. The moat around an AI product comes from what surrounds the model: data, workflow, distribution or a network. See AI moats.

Types of moats compared

Durable companies stack several moats, and most early startups pass few of these tests. That is normal: moats are built, not found. What matters early is a credible path to at least one.

How Moat Check scores the 10 moats

Moat Check researches existing competitors, names the incumbent best placed to copy the idea, then marks each test pass (10 points), partial (5) or fail (0), with one line of evidence. The total, out of 100, maps to a rating: No moat (0 to 29), Weak moat (30 to 54), Defensible (55 to 79) or Strong moat (80 to 100). A test passes only if it is true today or by design; an advantage that appears only "at scale" is a partial at best. The full rules are in the methodology.

Check your startup's moat

Questions