Comment by hinkley

This is why we don’t move data center load to the coldest available data center to reduce the AC power fraction of the cost. The cost of electricity is a significant fraction of the overall cost but not high enough to make up for stranded assets. Computers not running during their best years is expensive.

But I’m not sure that’s entirely correct, and maybe it’s time to revisit this.

Any system that is selling responsiveness as part of their service has to keep a certain amount of equipment sitting idle. That’s just how queuing theory works. So while you cannot move all server load to the coldest available zone, we should still be able to run that center near capacity and use the hottest one for all reserve capacity.

Power plants also have to deal with fines for exceeding emissions limits, but I suspect the problem here is that Bayesian analysis tells them that if a plant has to kick on early for some reason (early school release day, or another plant exceeded a maintenance window), it will still be needed for sure an hour from now, so it’s better to leave it running for 45 minutes doing nothing than to cycle it.