If it's 150 t/s, that's barely faster than Nvidia GPUs who are batching a lot more and are a lot more cost effective. Add in the Groq piece and Nvidia claims it can do 400 tokens/s.