Evaluating an AI Startup's Moat When Models Are Commoditizing

Every investor evaluating an AI startup eventually asks the same question: what stops someone else from doing this? The question has become harder to answer as powerful AI models become cheaper, more capable and widely available through a handful of large providers.

If every company can access similar models, the model itself is rarely the advantage. The moat has to come from somewhere else. This guide explains where defensibility really comes from in AI companies, how to test it in diligence, and what founders should show investors.

1. Why the Model Is Rarely the Moat

• Rapid improvement. Foundation models improve quickly, so an advantage built on a specific model or prompt technique can disappear with the next release.
• Falling costs. The cost of using capable models has dropped sharply, lowering barriers for competitors.
• Open models. Capable open-weight models give many companies strong alternatives to proprietary providers.
• Platform risk. Large model providers may build features that overlap with startups built on top of them.

For most application-layer companies, investors therefore look past the model to the business built around it.

2. Where AI Moats Actually Come From

• Proprietary data. Data that competitors cannot easily obtain, especially data generated by customers using the product, which improves results over time.
• Workflow depth. Products embedded in how customers work, with integrations into existing systems, are hard to replace even if a competitor has a similar model.
• Distribution. Existing customer relationships, channel partnerships or a strong brand can matter more than technology.
• Domain expertise and trust. In regulated fields such as healthcare, finance and law, accuracy, compliance and accountability create barriers general tools struggle to meet.
• Network effects. Products that become more valuable as more users join, such as marketplaces or shared data layers.
• Speed of execution. Not a permanent moat, but teams that ship and learn faster can stay ahead long enough to build other advantages.

Our comparison of vertical and horizontal SaaS explains why industry-focused AI companies often find these moats more easily.

3. How to Test a Moat in Diligence

1. The next-model test. If the next generation of foundation models is significantly better, does this company get stronger or weaker?
2. The replication test. How long and how much would it cost a well-funded competitor to rebuild the product?
3. The switching test. What would customers lose by switching? Look at integrations, stored data and retraining costs.
4. The data test. Is the company's data genuinely unique, and does it have the rights to use it?
5. The retention test. Do customers stay and expand? Net revenue retention is often the clearest evidence of a real moat.

4. What Founders Should Show Investors

• Why better models make your product better, not obsolete.
• What data you collect, how it improves results and who owns it.
• How deeply your product is embedded in customer workflows.
• Retention and expansion metrics that prove customers depend on you.
• How you manage dependence on any single model provider.

See our guides to how AI startups get funded and the AI wrapper question.

Frequently Asked Questions

Can an AI startup have a moat without its own model?

Yes. Most successful application-layer companies build moats through data, workflow integration, distribution and trust rather than proprietary models.

Is proprietary data always a moat?

Only if it is genuinely hard to obtain, improves the product meaningfully and the company has clear rights to use it.

What is the biggest risk for AI application companies?

That a foundation model provider or large incumbent builds the same capability directly into its products.

Which metric best shows an AI moat?

Strong net revenue retention, because it shows customers keep paying more even as alternatives appear.

The Bottom Line

As AI models commoditise, defensibility moves to what surrounds them: data, workflows, distribution, domain trust and network effects. Investors who test for these, and founders who can prove them, will separate durable AI companies from those that the next model release makes obsolete.

Global Capital Network connects AI founders with investors through our events and investor network. Get in touch if you are raising.

This article is general information, not investment advice.