Enterprise AI adoption constrained by data/IP loss risk; customers demand proprietary model control or closed-weight access to retain 'alpha'
“What slows down AI adoption in this country is people are saying, 'But I can't use these products cuz I'm not getting value, meaning I'm not using an application layer and I or I'm transferring the value of my business to someone else. Why would I do that?' ... Businesses don't like when they're paying too much and then they find out that somehow the actual insights of the business are being modeled by a far third party and sold to other people or they're losing their business.”
Why it matters
Karp is identifying a fundamental market friction: enterprise customers fear that using frontier model APIs exposes proprietary business logic to competitors via the model provider. This is NOT a cost issue but a strategic IP/competitive moat issue. If true at scale, it represents a ceiling on SaaS API adoption for frontier labs and creates demand for on-premise, customer-controlled alternatives.
Investment implication
This suggests demand for on-premise AI solutions, custom/fine-tuned models, and application-layer platforms that let enterprises retain data and model weights. Companies positioned to provide 'sovereign AI stacks' (Palantir, enterprise-focused open-source deployers) may see accelerated adoption; frontier labs may face customer churn or forced to shift to licensing/white-label models rather than API/token consumption.