Agreed to all of what you said, but also, performance in some circumstances is a discount. One of the Google Cloud employees said "Derek’s point about savings is about list price differences that result from total system performance (and not any sort of special discounting)."

I don't have any internal numbers, but imagine if Twitter could meet its needs with an N-node Redshift cluster or an N/2-node Dataproc cluster, because Dataproc is so much faster at flipping the bits or whatever Twitter wants to do (probably network performance really). Google could charge more per node and still come out cheaper because of performance.