Microsoft just announced that they are getting better price/performance ratios with AMD for inference. It's happening already. Most of GPU thirst for training a

Microsoft just announced that they are getting better price/performance ratios with AMD for inference. It's happening already. Most of GPU thirst for training at the moment but eventually most of workload will be inference. And that's depending on a big if, if users are willing to pay for LLM generated content as much as we're hyped about it right now.