Yes. We have a quad MI300A server and run several inference models on it. For $107k it has saved us so much money on tokens already and it's a heck of a lot faster than cloud services.

> Have you ever actually had anyone work with these chips? Developer ux on amd is terrible.

Just how much of dev ux do you need? A foundational library, of course, but as the AI companies keep saying, their models can vibe-code what's needed for those chips anyway.