Actually, Deploying local AI models is challenging for anyone because of the following major reasons.

1. High Costs – Requires expensive hardware (A100s, H100s) and skilled experts.

2. Responsibility & Risk – I've seen companies resisting in deploying local models becuase if the service goes down it's on them, but if they have API based models, they can blame the hosted companies, so everyone wants to play safe.

This makes API-based models a more convenient choice for most organizations.