The most common question we field from founders building AI companies is some variant of: "How do we build a moat when foundation models are getting better every six months and the underlying capabilities that our product is built on are commoditizing?" It is a genuine concern, and it deserves a more rigorous answer than the reassurance that "defensibility comes from execution" that circulates too freely in startup communities.
The honest answer is that not every AI application company has a strong durable moat, and founders who cannot articulate their moat thesis convincingly will struggle to raise their Series A in the current environment. Institutional investors have grown sophisticated about the model-commoditization risk, and hand-waving about product excellence and team quality is no longer sufficient.
Across our portfolio, we have observed three patterns that create genuine competitive defensibility in AI application companies: proprietary data assets, distribution advantages, and workflow depth. None of these is unique to AI -- all three appear in the moat analysis of non-AI software companies. But the specific mechanisms through which they create defensibility in AI applications are different enough that they warrant careful examination.
Moat Pattern One: Proprietary Data Assets
The most durable moat in AI application businesses is proprietary data that is either structurally unavailable to competitors or that would take years and enormous resources to replicate. The companies with this type of moat are not merely better at processing publicly available data; they have access to data that others cannot access, and that data advantage drives quantitatively better model performance that creates observable value for customers.
Proprietary data assets come from several sources in our portfolio. The first and most powerful is data generated by operating the product itself at scale. A company that has processed millions of specialized documents -- clinical records, legal filings, financial statements -- accumulates a training corpus that enables model performance improvements invisible to academic or public-data training approaches. Each new customer's data enriches the training set in ways that make the product better for all customers, creating a compounding advantage that scales with market adoption.
The second source is data accumulated through deep domain partnerships that create exclusive access to otherwise private datasets. Healthcare companies that have built IRB-compliant data sharing agreements with hospital networks, or financial services companies that have created consortium data arrangements with regulatory reporting bodies, have access to data assets that cannot be replicated through public data collection or synthetic data approaches.
The third source, which is often underappreciated, is behavioral data generated by human-in-the-loop workflows. AI products that incorporate expert human review and correction -- where physicians review AI-generated clinical summaries, or attorneys review AI-extracted contract provisions -- generate labeled correction data that creates a fine-tuning advantage. The human expert is effectively annotating the gap between the AI's current performance and the ideal output, and that annotation data trains the model to close the gap in ways that pure automated learning cannot replicate.
When evaluating data moats during diligence, we ask founders three questions that get to the heart of whether the data advantage is real: How much data do you have that competitors cannot access? How does more data make your product measurably better for customers, and what is the quantitative magnitude of that improvement? What prevents a well-funded competitor from replicating your data collection approach in 24 months?
Moat Pattern Two: Distribution Advantages
Distribution moats in AI businesses are fundamentally similar to distribution moats in conventional software businesses, but the specific channels through which they are created differ. In our portfolio, the most durable distribution advantages come from three sources: deep integrations with existing enterprise workflows, channel partnerships that create exclusive or preferred access to customer segments, and professional community ownership in domain-specific markets.
Deep workflow integrations create switching costs that compound over time. An AI product that is deeply embedded in a company's document management system, ERP, or industry-specific platform becomes extraordinarily difficult to replace not because of technical lock-in per se but because the cost of unwinding the integration, retraining users, and revalidating the workflows that depend on it exceeds the value of switching to any reasonable alternative. The best AI infrastructure investments in our portfolio are built around integration depth that is difficult for competitors to replicate even if their underlying model technology is comparable.
Channel partnerships that create preferred access to large customer segments represent a different type of distribution moat. A company that has built a formal partnership with a major professional services firm, a dominant vertical software platform, or a large insurance carrier can access thousands of potential enterprise customers through a single relationship. Building these partnerships requires significant investment in product compatibility, joint go-to-market development, and relationship maintenance, which creates a natural barrier to competitive entry even when the underlying technology is accessible to competitors.
Professional community ownership is the most underrated distribution moat in AI application businesses targeting specialized domains. A company that has built genuine brand equity and trust with a professional community -- emergency medicine physicians, commercial real estate attorneys, or manufacturing engineers -- gains access to informal word-of-mouth referral networks, community event platforms, and professional publication channels that are not available to companies that approach the market purely through conventional enterprise sales. Building community ownership requires sustained investment and authentic engagement over years, which is precisely why it creates durable competitive advantage: it cannot be acquired quickly by a well-funded competitor.
Moat Pattern Three: Workflow Depth
The third moat pattern is the one that is most specific to AI application businesses and the one that is most frequently underestimated during diligence: workflow depth. This describes the degree to which an AI product is embedded in the actual operating process of the customer organization, not just available as an auxiliary tool that could be replaced without disrupting core operations.
The distinction is important. An AI tool that is available to analysts but not integrated into approval workflows, reporting processes, or accountability structures is a feature, not a platform. It can be replaced with minimal disruption. An AI system that is embedded in the clinical protocol for emergency department triage, such that removing it would require restructuring the workflow that physicians and nurses have adapted to, is genuinely difficult to replace regardless of how good an alternative product might be.
Workflow depth creates switching costs that are qualitatively different from software switching costs because they involve human behavior change, not just technical migration. When an organization's staff have adapted their working patterns around an AI tool -- when the tool is generating artifacts that appear in official documents, compliance reports, or customer-facing outputs -- the cost of replacing it includes the full organizational cost of re-training staff, revalidating outputs, and managing the uncertainty that comes with change.
Building workflow depth deliberately requires founders to make product decisions that prioritize integration over ease of adoption. Products that are easy to try but shallow in their integration with customer workflows may generate high initial interest but will struggle to build the retention economics that enterprise software businesses require for sustainable growth. The best enterprise AI products we have backed make a deliberate choice to make their initial implementation more demanding in exchange for creating genuine workflow dependency that sustains high gross retention over time.
The Moats We Worry About
For completeness, it is worth describing the moat arguments that we hear frequently but do not find compelling during diligence, because understanding why certain arguments fail is as important as understanding why others succeed.
Brand and reputation in early markets. Being the first well-known company in a new AI category creates a brand advantage that can translate into distribution benefits. But brand is not a durable moat by itself -- it is only as strong as the product experience that sustains it. Companies that argue for brand moats without the underlying workflow depth, data advantage, or distribution mechanism to sustain those brand associations tend to have high initial growth followed by elevated churn as customers try alternatives.
Team quality and proprietary model architecture. Both of these arguments sound more compelling to technical founders than they are to experienced investors. Team quality is essential for execution but it is not a moat -- teams change, and talented engineers can be recruited by competitors. Proprietary model architectures may provide temporary performance advantages, but foundation model providers are improving rapidly and the window in which a proprietary architectural approach sustains a competitive edge tends to be shorter than founders expect.
Regulatory compliance as a moat. Some founders in regulated industries argue that their investment in regulatory compliance creates a moat because competitors would need to make the same investment to enter the market. This is partially true, but compliance is a barrier to entry, not a source of ongoing competitive advantage. Once competitors have achieved the same compliance certifications, the structural advantage disappears. Compliance is necessary for playing in regulated markets, but it is not sufficient for winning them.
Implications for Founders
If you are building an AI application company and cannot articulate a clear moat thesis in one of the three categories above, we would encourage you to spend time on this question before your Series A fundraise. The market has shifted, and institutional investors are specifically evaluating moat strength as a key dimension of Series A diligence in ways that were less common two years ago.
The good news is that if you have been operating in your market for 12 to 18 months, you probably have more moat-building progress than you realize -- data you have accumulated, workflow integrations you have built, community relationships you have developed. The work is often in recognizing and articulating what you already have, and then being deliberate about accelerating the moat-building investments that are most likely to create durable competitive advantage in your specific market.
Jordan Webb is a General Partner at Orbit AI. He previously led ML platform engineering at a major social media company and served as CTO at two venture-backed AI companies. He writes about AI architecture, competitive strategy in AI applications, and the technical dimensions of building defensible AI businesses. This article represents his personal views.