Comment by moondistance

Yes, for many applications.

Meta, OpenAI, Crusoe, and xAI recently announced large purchases of MI300 chips for inference.

MI400, which will be available next year, also looks to be at least on par with Nvidia's roadmap.