Running AI models on infrastructure an organisation controls so its data stays in-house. Covers self-hosted open-weight models, on-premises and private cloud GPU platforms, RAG over internal data, access control, and costs against public AI services.