Does anybody actually work with H100s and the like? Their failure rate is so high, I dont understand why anybody will even consider it feasible to put the machines in orbit or even the sea. By my ballpark estimate, if you have 800 H100s, after 6 months, about 100 would be overheating or throttling, and a few will disappear and one or two will crash the machine with load.

> Does anybody actually work with H100s and the like?

They don't. The expectation the cloud develops in people is that magic computers just appear. They're living at a virtualized layer where all the nitty gritty of real machines going down and needing to be serviced all the time is handled by unseen minions (sorry SREs and DC staff) and cluster management and provisioning software.

The reality is that datacenters in space is mind-boggling stupid, just from the infeasibility of maintenance alone.