Defensibility and Moats in AI: A Cambridge Angel Perspective
"How do I prove defensibility?" This is a question Cambridge Angels hears from a number of AI founders. In a landscape where today's breakthrough risks becoming tomorrow's commodity, Cambridge Angels member Chris Mitchell has helpfully provided his views on what separates fundable AI ventures from the rest of the pack.
Cambridge Angels evaluates AI opportunities by looking beyond the hype to identify sustainable competitive advantages. In an era of rapid technological shifts, defining and demonstrating a "moat" is critical for long-term viability.
1. AI Definition: Beyond LLMs to the Broader Field
When Cambridge Angels discusses AI, the definition extends far beyond Large Language Models (LLMs). While generative AI and LLMs are current focal points, "defensible" AI often involves the broader field, including physical AI, computer vision, and specialized machine learning. For instance, some companies utilize computational physics and proprietary algorithms for post-RGB machine vision, representing a more specialized application of AI than general-purpose text generation. Similarly, other firms focus on transforming unstructured medical records into validated legal insight applying AI to complex, regulated domains.
2. Speed of Development and Breakthrough Vulnerability
The speed of development in AI means that today’s breakthrough can quickly become tomorrow’s commodity. Moats are dynamic; they can evaporate as technology breakthroughs happen. A key risk is the "Good Enough" competitor: an enterprise might prefer a 70% effective, easily deployable LLM-based solution over a complex, 100% effective solver-based engine. Investors look for companies that can maintain a lead despite the rapid pace of the industry, often through continuous innovation or deep technical integration that is not easily replicated by general updates to foundational models.
3. Categories of Moats and Defensiveness
Defensiveness in AI can fall into several categories:
Proprietary Data: A library of well-characterized, unique data that competitors cannot easily access is a primary moat.
Technical/IP Moats: Patents on specific approaches, such as novel approaches to Quantum Key Distribution, provide legal barriers to entry.
Hardware-Software Co-design: Deep integration between novel hardware (like metasurfaces) and proprietary algorithms creates a complex barrier that software-only players cannot easily hurdle.
Regulatory/Clinical Validation: In sectors like healthcare, having a platform that is already approved for sensitive patient data and clinically validated creates a significant "beachhead".
4. Evolution of Moats Over Time
A company’s moat typically evolves as it matures. Initially, the moat may be purely technical or IP-based - the "secret sauce" of a research spinout for example. As the company scales, the value can often shift toward the data accumulated through operations and the deep integration into customer workflows. For example, a company providing autonomous enterprise operations might start with a unique technical solver, but its long-term defensibility will come from the "world-model" it builds using specific customer data, which becomes increasingly difficult for a new entrant to replace.
5. Demonstrating a moat
To effectively demonstrate a moat, founders must provide concrete evidence of sustainable competitive advantages. This includes technical performance benchmarks that showcase superior capabilities, robust customer validation such as successful paid pilots with major industrial firms, and where appropriate a clear, well-defined IP strategy. Ultimately, proving a moat requires showing a "decisive competitive edge"—the specific, technical, or operational reason why even a well-funded competitor cannot easily displace the solution.