Storj has certain node-selection criteria (like separate /24 nodes storing pieces) to reduce the chance too many pieces maybe be on hardware that can fail at the same time. But ultimately nobody can tell if nodes are on independent systems or not. Like two nodes using IPs that geolocate to opposite sides of the globe… could be in two docker containers right next to each other.
And there are exceptions: like SNOs participating in Select don’t have the /24 limit applied. But when you apply for Select they make sure you’re running more reliable configs.
Obviously unreliable nodes can still fail audits and become permanently disqualified. But if they are reliable… there’s no reason large SNOs can’t get good chunks of all the capacity-reservation data being sent now. The repair system has been working very well: I don’t think any customer data has been lost - and they trust it enough it should handle an entire country dropping off the Internet. One SNO going offline… even an large one… is no big deal.