♟️ AI Strategy
Definition: The alignment of a company’s artificial intelligence capabilities with its broader business objectives to create sustainable competitive advantage and defensibility (moats).
A central topic in modern strategic management courses at HBS and Wharton
🔑 The Core Problem: The “Thin Wrapper”
If every company has access to the exact same foundational models (GPT-4, Claude, Gemini) via APIs, how do you build a competitive advantage?
If a startup simply connects an interface to an OpenAI API (a “thin wrapper”), it has no defensible moat. The moment the foundational model updates its native interface to include that feature, the startup’s entire value proposition vanishes (often called being “Sherlocked”).
🏰 Building Moats in the AI Era
To build a defensible AI strategy, companies must rely on advantages outside of the model itself.
1. The Proprietary Data Moat
- Data Network Effects: More users → more interaction data → better fine-tuning/RLHF (Reinforcement Learning from Human Feedback) → better product → more users.
- Unique Corpora: A law firm possessing 50 years of proprietary, highly confidential contract negotiations has an unassailable data advantage to fine-tune an AI that no startup can replicate.
2. The Distribution Moat
- Incumbent Advantage: Microsoft integrating AI into Office 365 or Salesforce adding Einstein to its CRM. The AI might not be the absolute best, but the fact that it is smoothly integrated into the daily workflow of 100 million users beats a standalone competitor.
- “Will the incumbents get AI before the startups get distribution?“
3. The Workflow / UX Moat
- AI that seamlessly fits into deeply complex, specialized workflows.
- E.g., Harvey (legal AI) or proprietary medical diagnostic tools. The moat is the UI, compliance, and deep understanding of the user’s daily friction points, not just the text generation.
4. The Hardware / Compute Moat
- Securing massive GPU capacity (H100s) at scale or developing proprietary silicon (e.g., Google’s TPUs).
📉 Three Tiers of AI Strategy
| Tier | Approach | Risk Level | Value Capture |
|---|---|---|---|
| Foundation Providers | Building the massive base models (OpenAI, Anthropic, Google) | Extremely High (Billions) | Massive |
| Model Fine-tuners | Taking open-source models (Llama) and training them for deep niche tasks | Moderate | High (Niche Monopolies) |
| Integrators | Building applications using APIs | Low | Lowest (Commoditized) |
🔗 Connected Concepts
- Competitive Advantage — The traditional framing of moats
- Network Effects — How data flywheels create lock-in
- Large Language Models — The underlying technology
- Disruptive Innovation — How AI is upending traditional software
← 📉 Data & Analytics MOC | Related: Competitive Advantage · Large Language Models