The 'Intelligence Stack' Becomes AI's New Moat
As frontier AI models become commoditized, a new analysis argues that competitive advantage is shifting from raw model quality to the "intelligence stack." This refers to how well a product integrates, orchestrates, and applies AI models in specific user contexts, with the edge coming from superior data, UX, and network effects.
- The trend towards commoditization is driven by the proliferation of powerful open-weight models from entities like Mistral and the release of highly efficient models like DeepSeek, which have dramatically lowered the cost of and access to high-quality AI reasoning. - As base models become commodities, value is shifting to the "plumbing" and "fixtures"—the tooling, data pipelines, and domain-specific applications that use the models. This moves the competitive focus from building the largest model to orchestrating the right model for a specific task. - A key component of the intelligence stack is proprietary first-party data; for example, the company Tempus AI leverages its unique, large-scale dataset of de-identified clinical and molecular data to train AI models for use in precision oncology. - AI products are developing "data network effects," where the system improves automatically with use. Every user interaction, such as a prompt, click, or edit, provides feedback data that refines the AI, creating a virtuous cycle where more usage leads to a better product, attracting more users. - Deep workflow integration is another defensible moat; by embedding AI so deeply into a customer's operational processes, it becomes indispensable. For instance, the AI coding assistant Cursor, from Anysphere, has seen rapid adoption by integrating directly into the engineering workflow and reportedly signed $100 million in contracts in about a year. - The rise of the intelligence stack has shifted user expectations, especially in business applications. While early AI tools that simply provided answers were novel, customers now expect systems that can reliably execute complex, end-to-end work without supervision. - Companies are increasingly adopting a "booster" or "builder" strategy rather than just being a "buyer" of off-the-shelf AI. This involves integrating available models with proprietary data ("booster") or building custom models on a strong digital core to create a unique competitive advantage.
