I don’t think that’s necessarily incompatible with what I suggested. They could just leave the backup servers offline until they’re actually needed which shouldn’t cost them anything (or at least not much; some cloud providers charge for a VM’s storage usage regardless).

Assuming that Signal’s servers were designed by competent engineers, the engineering cost to make a change like this shouldn’t be that bad. Though judging by Whittaker’s comments, that may be a bad assumption.