Nvidia will either be selling inference on its own proprietary foundation model (e.g. like OAI) or will release a SOTA open source LLM that competes with the SOTA foundation lab proprietary models (e.g. Meta’s LLama 3 400b).