There's also the availability factor. OpenAI has been known to go down on occasion and without warning. If a product relies on an LLM, I wouldn't feel great about the observed uptime of OpenAI APIs.
This doesn't take away from high availability being a legitimate need to host your own LLM, though.