Comment by dwroberts

Is this admission that google’s proprietary chips etc. are not cutting it? Why would you need a bunch of nvidia GPUs if you have your own silicon? (AFAIK they have their own for both inference and training do they not?)