You know that many of us here are HPC folks who know the nuances... and simply disagree? I mean, sure, I don't think Intel should have taken the contract or gotten any positive PR for this, but at the fundamental level, many people on this site did supercomputing and HPC at national labs or universities, and now work on machine learning HPC on the cloud. My experience spanning both makes me think that chasing time on the fastest supercomputers is not the best way for scientists to be productive.