Nvidia already have a plan in place for fast inference. They acquihire Groq last year