Running large language models at the enterprise level often means sending prompts and data to a managed service in the cloud, much like with consumer use cases. This has worked in the past because ...