Which layer of the AI stack is most likely to capture disproportionate value as models commoditize?

The layer most likely to capture disproportionate value as models commoditize is the Application Layer, specifically those that integrate into high-friction enterprise workflows or proprietary data loops.

As base models become cheaper and more ubiquitous (commoditization), the competitive advantage shifts away from raw intelligence toward:

• Vertical Integration: Applications that solve specific, non-trivial problems (e.g., healthcare diagnostics, legal compliance, or industrial automation) create "moats" through specialized domain knowledge.
• Data Flywheels: Platforms that leverage unique, proprietary data to fine-tune or augment models create outcomes that generic, commoditized models cannot replicate.
• Workflow Embedding: Products that become the "system of record" or the primary interface for an organization’s work capture value by being difficult to displace, regardless of which underlying model they utilize.

While the Infrastructure layer (compute and hardware) captures value through scarcity, and the Model layer currently captures value through R&D investment, the Application layer captures value through stickiness and end-user utility. When intelligence is cheap, the value lies in how it is applied, not in the intelligence itself.