Interesting, but somewhat incomplete analysis IMO. If you are running a billion + dollar budget for compute spending, you’re surely going to negotiate pricing

Interesting, but somewhat incomplete analysis IMO. If you are running a billion + dollar budget for compute spending, you’re surely going to negotiate pricing with your vendor, and while that won’t bring things to parity, it will bring them much closer together. If you are spending this kind of money, you’re likely doing a lot to get your workloads into some reasonably geographically aligned areas, and if you’re peering with other services, they tend to be running on AWS, which means choosing another provider can significantly increase those latency costs. While we’re on the subject of bandwidth, what do you think data transfer pricing will be between your other cloud provider and this high compute instance you picked up on the cheap? Odds are they will more than negate your cost savings, and again at a latency cost. Let’s say you really want to save huge amounts of money though. The easy answer is probably in moving to a spot instance targeted architecture where you can typically buy the same server from AWS for less than 20 percent of the on demand price. You can always fall back to on demand when they are not available.