AI and Competitive Moats: Valuing Proprietary AI Systems

The most important question in AI valuation is not "Does this company use AI?" but "Can a competitor replicate this AI capability, and how long would it take?" The answer determines whether AI creates a durable competitive moat — a structural advantage that protects margins and market position over time — or a temporary feature advantage that evaporates as soon as competitors adopt similar technology.

In a market where AI capability commands significant valuation premiums, understanding the defensibility of that capability is worth millions in better investment decisions.

The Four Types of AI Moats

1. The Data Moat

The most durable AI moat is built on proprietary data that improves the model's performance and cannot be replicated by competitors. The data moat has three dimensions:

Exclusivity: Is the training data exclusively available to this company? Data from proprietary customer interactions, operational processes, or sensor networks is exclusive. Data from public sources, purchased datasets, or scraped web content is not.

Scale: Does the company have more relevant data than any competitor could feasibly accumulate? Scale advantages compound — a model trained on 10 million transactions outperforms one trained on 100,000, and the gap widens with each additional data point.

Freshness: Is new data continuously generated by the business's normal operations? A company whose customers generate training data through their daily use of the product has a self-reinforcing data advantage.

2. The Model Moat

Some AI applications require deep domain expertise embodied in custom model architectures, novel training techniques, or unique algorithmic approaches. The model moat exists when:

• The model architecture reflects proprietary research that advances beyond publicly available methods
• The training process incorporates domain knowledge that general-purpose approaches cannot replicate
• The model's performance depends on engineering decisions that require years of iteration to discover

Model moats are less durable than data moats because algorithmic advances eventually become public through research papers, open-source implementations, and talent mobility. A model moat typically provides 12-36 months of advantage before competitors close the gap.

3. The Integration Moat

AI that is deeply embedded in business processes, customer workflows, and partner systems creates switching costs that competitors cannot overcome with better technology alone. The integration moat exists when:

• Removing the AI system would require rebuilding core business processes
• Customers have adapted their workflows around the AI system's outputs
• The AI system is interconnected with other systems in ways that create network dependencies

4. The Network Moat

The most powerful AI moat emerges when each additional user improves the AI system for all users, creating a self-reinforcing network effect. Examples include:

• Recommendation engines that improve with more user behaviour data
• Marketplace matching algorithms that improve with more participants
• Collaborative filtering systems that become more accurate with more users

Network moats are exponentially defensible because the value gap between the incumbent and any new entrant widens with each new user. They are also the rarest — most AI systems do not benefit from true network effects.

Valuing AI Moats

The moat assessment matrix

Practical valuation approach

To value an AI competitive moat, use the relief from royalty method adapted for AI assets:

1. Estimate the revenue attributable to the AI moat — what revenue would decline or disappear if the AI advantage were eliminated?
2. Apply a royalty rate reflecting what a licensee would pay for equivalent capability (typically 3-8% for AI-driven features)
3. Project over the estimated moat duration — how long before competitors can replicate?
4. Discount at an appropriate rate reflecting the uncertainty of AI asset value (typically 15-25% for AI-specific assets)

Implications for Investors

When evaluating AI-enabled intangible assets in investment contexts, three principles apply:

Price for the moat, not the model. Two companies with identical AI models have very different values if one has exclusive data and deep integration while the other has public data and loose coupling.

Assess moat trajectory. Is the moat strengthening or weakening? Companies with data flywheels and network effects have strengthening moats. Companies dependent on model innovation have moats that erode as competitors catch up.

Verify moat claims. Like AI washing, "moat washing" is common. Companies claim data exclusivity when their data is assembled from public sources, or claim network effects when users do not actually improve the system for each other. Independent verification is essential.

The Opagio questionnaire and Growth Platform assess technology competitive advantage systematically, including data asset defensibility, model differentiation, and integration depth.

Ivan Gowan is Founder and CEO of Opagio. He spent 15 years building and defending technology competitive advantages at IG Group (LSE: IGG), where platform technology was the primary source of competitive differentiation. Learn more about the Opagio team.