The 7 Most Powerful Moats For AI Startups
Y CombinatorKey Moments
AI startups can build moats beyond speed using 7 powers: process, cornered resources, switching costs, counterpositioning, branding, network effects, and economies of scale.
Key Insights
Speed is the initial moat for AI startups, but deeper defenses are needed for long-term success.
The seven powers framework, originally for traditional businesses, is still relevant for AI startups.
Process power, built on complex, honed AI systems and specialized knowledge, creates defensibility.
Cornered resources can include patents, regulatory approvals, or unique access to data and customer understanding.
Switching costs in AI can arise from deep integration, custom workflows, and long onboarding, alongside AI's ability to reduce old-school data migration costs.
Counterpositioning involves competing in a way that incumbents cannot easily replicate due to cannibalization fears, often leveraging a superior product or different business model.
Network effects in AI are driven by data; more usage leads to better models, creating a virtuous cycle.
Economies of scale are primarily relevant at the foundation model layer due to high training costs, with application-level scale economies being less common but emerging in areas like web crawling.
Founders should prioritize solving a painful customer problem first (speed) before obsessing over long-term moats.
THE IMPORTANCE OF MOATS IN THE AI ERA
The discussion centers on the critical need for defensible moats in AI startups, which go beyond initial speed. While speed is essential for early traction, a sustainable business requires long-term advantages against inevitable competition. The conversation highlights a potential misconception among founders, particularly in AI, about the difficulty of building moats, often fueled by the 'ChatGPT rapper' meme suggesting easy replication. The core argument is that deep and interesting moats do exist for AI companies, following timeless business principles.
UNDERSTANDING HAMILTON HELMER'S SEVEN POWERS
The foundation of the discussion is Hamilton Helmer's 'Seven Powers' framework, a business strategy concept originally outlined in 2016. Although the book's examples are from the pre-AI internet era, the underlying principles remain highly relevant. The 'powers' are essentially categories of moats that protect a business from competition. The framework is presented not as a rigid set of rules, but as timeless strategies that can be adapted to the unique landscape of AI startups, providing founders with a structured way to think about long-term defensibility.
SPEED: THE INITIAL MOAT AND BEYOND
Speed is unequivocally identified as the primary moat for early-stage startups. In the nascent AI space, the ability to execute rapidly and ship features daily, as exemplified by Cursor, allows startups to outmaneuver larger, more bureaucratic incumbents. This speed is crucial for iterating, finding product-market fit, and establishing a strong initial user base. However, once a company gains traction and proves its value, it must then focus on developing deeper, more sustainable moats from the seven powers framework to ensure long-term survival and growth.
PROCESS POWER: COMPLEXITY AND EXECUTION
Process power derives from building a complex, finely-honed operational system that is difficult for competitors to replicate. In AI, this translates to highly optimized AI agents that excel in real-world conditions, often built through years of iterative improvement and specialized knowledge. Examples like Case Text, Greenlight (KYC for banks), and Casa (loan origination) illustrate this. The hackathon version of these tools is easily built, but the true defensibility lies in achieving high reliability, accuracy, and specialized understanding for mission-critical applications, requiring painstaking effort beyond initial concepts.
CORNERED RESOURCES: UNIQUE ACCESS AND PATENTS
Cornered resources represent assets that are independently valuable and not easily arbitraged or replicated. This can include patents, regulatory approvals, or exclusive access to crucial inputs. In the AI context, this extends to securing special access or partnerships, such as Scale AI and Palantir working with the DoD, which involves significant investment in infrastructure (like SKIFs) and building trust. It can also mean capturing the 'brain space' within key customer organizations or leveraging proprietary data and workflows obtained by being 'forward-deployed' with clients.
SWITCHING COSTS: LOCKED-IN CUSTOMERS
Switching costs create defensibility by making it prohibitively expensive or difficult for customers to move to a competitor. Traditionally, this involved data migration and retraining (e.g., Oracle databases, Salesforce). For AI, while LLMs can potentially lower old-school data migration costs, new switching costs emerge from deep integrations and lengthy onboarding processes with enterprises, custom tailoring AI agents to specific workflows (e.g., Happy Robot, Salient). Consumer AI also sees switching costs develop through personalization and AI memory, making it harder for users to leave a familiar, personalized interface.
COUNTERPOSITIONING: AVOIDING CANNIBALIZATION
Counterpositioning involves adopting strategies that incumbents cannot easily copy because it would undermine their existing business. A key example is the pricing model conflict: many incumbents charge per seat, but successful AI automation reduces the need for employees, thus reducing revenue. Newer AI startups often use work- or task-based pricing, pushing for actual task completion. This also highlights how incumbents struggle to reset engineering cultures to embrace AI. Furthermore, second-movers can counter-position by focusing on application layers or superior out-of-the-box functionality, as seen with Legora versus Harvey or Gig ML versus established players.
BRANDING: ESTABLISHING RECOGNITION
While brand as a moat takes time to build, its power is undeniable. The example of OpenAI's ChatGPT surpassing Google's Gemini in consumer usage, despite Google’s massive user base and advanced models, illustrates this. OpenAI built a strong consumer brand as the primary AI application, demonstrating that even with technological parity or superiority from an incumbent, a strong brand identity can win consumer preference. This is closely related to counterpositioning, where a new brand can carve out a niche by offering a clearly differentiated value proposition.
NETWORK EFFECTS: DRIVEN BY DATA AND USAGE
In the AI era, network effects are primarily driven by data. The more users interact with an AI product, the more data is generated, which can be used to train better models. These improved models, in turn, attract more users, creating a virtuous cycle. Examples include ChatGPT feeding user data into future model training and Cursor using developer keystrokes to enhance its autocomplete. For enterprise AI, access to private customer data through forward-deployed engineers enhances model performance, further strengthening the network effect through continuous feedback loops and evaluations (evals).
ECONOMIES OF SCALE: MODEL TRAINING AND INFRASTRUCTURE
Economies of scale are most powerful at the foundation model layer, where training state-of-the-art LLMs is extremely capital-intensive, limiting the number of players. However, the cost of inference (using the model) can be made relatively inexpensive once the initial investment is made. While this moat is strong for large AI labs, it's less common at the application layer. Emerging examples include companies like Exa, which invest heavily in web crawling infrastructure to provide data access for AI agents, creating a scale advantage through a reusable, expensive fixed asset.
RETHINKING AI STARTUP STRATEGY
The overarching advice for AI startup founders is to not get bogged down in prematurely assessing long-term moats. The most crucial first step is identifying a significant, painful customer problem and solving it with speed and relentless execution. Once a business gains traction, then the focus can shift to building more enduring moats. The framework should guide development, not pre-emptively disqualify ideas, emphasizing that problems worth solving will naturally lead to opportunities for building defensible positions over time.
Mentioned in This Episode
●Software & Apps
●Companies
●Organizations
●Books
●Concepts
●People Referenced
Building Moats for AI Startups: Key Strategies
Practical takeaways from this episode
Do This
Avoid This
Common Questions
A business moat is a defensive advantage that protects a company from competition. For AI startups, understanding moats is crucial to building sustainable, long-term businesses rather than short-lived ventures easily replicated or outcompeted by larger players.
Topics
Mentioned in this video
Discussed in the context of cornered resources, working with the DoD on AI initiatives.
Mentioned for his perspective on startups as 'forward deployed engineering teams for the labs'.
A YC startup developing customer support software for HVAC companies, demonstrating how AI startups can capture higher wallet share by focusing on critical workflows.
A business moat derived from exclusive access to valuable assets like patents, regulatory approvals, or proprietary data.
An earlier payment processor that Stripe came after.
An AI agent company for loan origination in banks, showcasing process power and defensibility due to its complex, mission-critical operations.
An AI voice agent company for the financial industry, building custom workflows for banks.
Mentioned as a defensible SaaS company with a moat built on extensive software and deep backend logic.
An earlier payment processor that Stripe came after.
A book by Hamilton Helmer that outlines seven categories of business moats, originally focused on internet companies but applicable to AI startups.
A business moat characterized by complex, hard-to-replicate systems and operations.
A moat created when customers face significant financial or operational hurdles to switch to a competitor.
Heating, Ventilation, and Air Conditioning, a sector where vertical AI SaaS companies like Aoka are finding significant growth potential.
A language learning app using LLMs and voice to help users practice and learn, counterpositioning Duolingo by focusing on actual speaking practice.
A recent YC batch company building plays similar to Exa, crawling the web for AI agents.
Department of Defense, mentioned in the context of Scale AI's work and the procurement process for government contracts.
An earlier food delivery service that DoorDash came after.
Author of 'The Seven Powers', who taught at Stanford Economics School.
A well-known customer support company that Giga ML competes against.
Highlighted as the primary and initial moat for AI startups, emphasizing relentless execution and rapid product development.
A strategy where a company competes by doing something that would cannibalize the incumbent's business if they tried to copy it.
A moat achieved through massive investment that lowers the cost per unit, particularly relevant in AI model training.
Customer of Happy Robot, highlighting how deep integration into specific workflows creates switching costs.
An ATS (Applicant Tracking System) mentioned as an example of traditional SaaS with high switching costs related to data migration.
A moat where the value of a product increases with more users, often driven by data in the AI era.
Mentioned as someone who shared details about Cursor's rapid product development cycle.
An AI company integrating deeply into logistics workflows for companies like DHL, creating switching costs.
A recent YC batch company building plays similar to Exa, crawling the web for AI agents.
Well-known customer support company that Giga ML is competing against.
A company providing search for AI agents via an API, requiring significant investment in web crawling, which forms its economies of scale moat.
More from Y Combinator
View all 100 summaries
54 minThe Future Of Brain-Computer Interfaces
38 minCommon Mistakes With Vibe Coded Websites
20 minThe Powerful Alternative To Fine-Tuning
24 minThe AI Agent Economy Is Here
Found this useful? Build your knowledge library
Get AI-powered summaries of any YouTube video, podcast, or article in seconds. Save them to your personal pods and access them anytime.
Try Summify free