Where Value Accrues in AI-Native Businesses
The layer of the AI stack most likely to produce durable enterprise value is not the layer most often discussed.
The visible and the durable
Attention accrues to the visible layer — models, interfaces, benchmarks. Value tends to accrue elsewhere — in proprietary workflows, in embedded distribution, in the compounding advantage of a system that gets sharper every time it is used. The two rarely occupy the same company.
What we look for
Businesses where the AI is a means, not the identity. Where the moat is not the model but the specific problem, the specific data flywheel, the specific relationship with the customer that a general system cannot replicate. Where switching costs are real, not asserted.
“The moat is rarely the model. The moat is the specific relationship the system has with the specific problem it solves.”
What we discount
Feature-level differentiation on top of foundation models that any capable team could rebuild in a quarter. Distribution strategies that assume permanent access to a platform's user base. Cost structures that only work while inference is subsidised.
You may also find helpful