[Narew]

It's for inference only not training. In this use case, there is lots of device that's not Nvidia. For server you have Google tpu, for more close to public there is the Apple Neural Engine for example.