How to Assess Whether an AI Company Really Has a Moat

Artificial intelligence may create enormous economic value without every AI company becoming an attractive investment.

That distinction matters when investors assess highly valued private developers. A capable model, impressive growth or prominent customers can demonstrate commercial potential. None of these automatically proves that a business has a durable moat.

The central question is not simply whether the technology is good. It is whether the company can retain customers, defend its margins and earn an attractive return on the capital required to compete.

Start by defining the moat

A moat is a structural advantage that makes it difficult for competitors to take customers or copy a company’s economics.

For an AI developer, several possible sources of advantage deserve investigation:

• Distribution: Does the company control a widely used platform or route to customers?
• Switching costs: Would changing provider disrupt important systems, workflows or data connections?
• Proprietary data: Does it have lawful access to valuable information that rivals cannot easily reproduce?
• Brand and trust: Do customers see the provider as safer or more dependable for sensitive work?
• Cost advantages: Can it deliver comparable results more cheaply than competitors?
• Ecosystem strength: Have developers, consultants and software partners built products around its technology?

Investors should be precise here. Saying that a company has strong technology is not the same as identifying a moat. Technology can be copied, improved upon or made cheaper. A genuine moat should influence customer behaviour and financial outcomes.

Test the company’s pricing power

Pricing power is one of the clearest signs of competitive advantage. A company has it when customers accept higher prices, or resist moving to cheaper alternatives, because the product delivers distinctive value.

The first question is whether customers are buying access to a model or a complete business solution.

Basic model access may be vulnerable to price competition if several providers produce broadly acceptable results. A deeply integrated product can be more defensible. If it is connected to internal databases, compliance processes and staff workflows, replacement may become expensive and risky.

Investors should look for evidence that revenue comes from more than promotional pricing or short-term experimentation. Useful questions include:

1. Do customers expand their spending over time?
2. How easily can they use a competing model?
3. Are contracts long enough to provide revenue visibility?
4. Does the product save enough time or money to justify its cost?
5. Would lower industry prices damage margins or stimulate enough extra demand to compensate?

Pricing power does not always mean charging the highest price. A provider may deliberately charge less because its cost base is lower. What matters is the gap between customer value and the cost of delivering the service.

Examine switching costs carefully

Technology companies often describe integrations as evidence of customer loyalty. Investors should ask whether those integrations create real switching costs or merely temporary inconvenience.

A customer may be able to place several AI models behind the same interface and route each task to the cheapest suitable option. If so, the underlying providers could lose bargaining power even while AI usage rises.

By contrast, switching costs may be meaningful where changing provider requires extensive testing, staff retraining, regulatory approval or redevelopment of important applications. Reliability and security may also matter more than small price differences in critical workflows.

The practical test is simple: what would a rational customer lose by moving elsewhere?

If the answer is little more than a few days of engineering work, the moat may be weaker than it appears.

Do not ignore capital intensity

Rapid revenue growth can look less attractive when it requires continual spending on computing infrastructure, specialist staff and model development.

Investors therefore need to distinguish accounting profit from economic returns. A business may report improving operating performance while still needing substantial additional investment to remain competitive.

Consider four questions:

• How much capital is required to serve each additional customer?
• Does the cost per task decline as usage grows?
• Must the company fund repeated development cycles simply to keep pace?
• Who captures most of the value - the model developer, infrastructure supplier, distributor or end customer?

The last question is especially important. A fast-growing industry can produce disappointing returns for companies caught between powerful suppliers and price-sensitive customers.

This principle applies beyond AI. My Samsung Electronics financial statements analysis provides a separate example of why investors should examine investment requirements alongside headline business performance.

Watch for commoditisation

A product becomes commoditised when customers see competing versions as sufficiently interchangeable. Competition then tends to shift towards price, availability and distribution.

Investors should monitor the direction of travel rather than trying to declare that all AI models are either unique or identical. Different parts of the market may develop differently.

Signs of rising commoditisation risk could include:

• Customers routinely using multiple providers
• Falling prices for similar levels of performance
• Software making it easier to switch between models
• Buyers caring more about cost than model identity
• Open or lower-cost alternatives becoming adequate for common tasks

Possible defences include specialisation, superior reliability, exclusive data, trusted deployment tools and ownership of the customer relationship.

The best model does not necessarily become the best business. Distribution, convenience and integration can matter just as much as technical performance.

Treat private valuations with caution

Valuing a private AI company is difficult because disclosure may be limited and transactions can involve different share classes, investor rights or strategic considerations.

A quoted valuation should not be treated as a precise measure of what every share is worth. Investors should instead ask what assumptions the valuation requires.

A sensible framework might consider:

• The long-term revenue opportunity
• Sustainable gross margins
• Future capital requirements
• Competitive intensity
• Customer concentration
• Dilution from additional fundraising
• The probability of achieving durable free cash flow

Scenario analysis is more useful than a single heroic forecast. Build an optimistic case, a middle case and a difficult case. The difficult case should include lower prices, higher computing costs and weaker customer retention.

If the valuation only looks reasonable under the most favourable scenario, the margin of safety is thin.

A practical risk-control checklist

Retail investors cannot eliminate uncertainty, but they can control how much they pay and how much exposure they take.

Before investing in any company linked to a fashionable technology theme, ask:

1. Can I explain the moat without relying on technical superlatives?
2. Is customer loyalty proven or merely assumed?
3. Are margins likely to survive stronger competition?
4. How much reinvestment is required?
5. What happens if the core product becomes cheaper and more interchangeable?
6. Does the valuation allow for setbacks?
7. Am I analysing a business or reacting to a compelling story?

Investors building a broader process may also find my UK investing guide useful.

The bottom line

AI may remain strategically important while individual developers struggle to defend pricing and earn attractive returns. Industry growth and shareholder returns are not the same thing.

A durable AI investment case needs more than a powerful model. It requires customer dependence, disciplined spending, defensible margins and a valuation that reflects uncertainty.

The most useful question is not whether an AI company will win the technological race. It is whether lasting competitive advantages will allow its owners to capture enough of the value created.