Cassandra, for instance, provides an easy snapshot API, and you can back up the snapshots using whatever infrastructure you'd normally use to back up TB of data. (If you don't have such an infrastructure, yeah, that's a problem, but not Cassandra's fault. :)
Edit: I should point out that since Cassandra supports multiple data center replication already, I'm having trouble picturing a scenario where you want to do anything w/ the snapshots besides just leave them on the Cassandra nodes themselves (plan extra HDD capacity as necessary depending on how long you want to keep them). But some such scenarios probably do exist.