#650
The AI Landlord Problem

The most valuable thing a serious enterprise owns is the proprietary data and business logic that rivals cannot copy, which is why control over it has become the defining anxiety of the AI moment. Inside a bank, a hospital, or a defense contractor, a raw large language model flips from asset to liability, because without a control layer on top, everything it sees travels upstream to a server the customer does not own, caching the data and absorbing the logic that makes its user valuable along the way.

Alex Karp, CEO and co-founder of Palantir, recently put that discomfort into words, arguing that critical institutions never run these models unguarded and that trust between enterprises and the labs has broken down over one question: who ends up owning the data, the weights, and the advantage. The pricing only sharpens the doubt, because if the technology were as transformative as advertised, wouldn't its makers want a share of the value they create rather than meter it by the token? Charging a steep toll while quietly harvesting a client's alpha is a hard thing to sell.

The response now spreading through the most sophisticated buyers is to stop renting and start owning: keep the data in-house, lean on open-weight models they can host and inspect, and refuse to wire everything to one lab's roadmap. The point underneath is simple, which is that the moat was never the model, it was everything the customer already had. A bit like (or not at all 😂) Rice, Saka, Eze, and Madueke slipping one past their Arsenal captain Ødegaard tonight in Miami.