I don’t know if it is the „correct“ way but I do it the other way around. I have a server and a backup server. Server user can‘t even see backup server but packs a backup, backup server pulls the data with read only access, main server deletes backup, done.
For an organisation hosting as many companies data as this one I'd expect automated tape at a minimum. Of course, if the attacker had the time to start messing with the tape that's lost as well but it's unlikely.
It depends what's the pricing. For example ovh didn't keep any extra backup when their datacenter took fire. But if a customer paid for backup, it was kept off-site and was recovered
It might be even pretending to be a big hosting company when they're actually renting a dozen deds from a big player, much cheaper than maintaining a data center with 99.999% uptime
Fundamentally there's no need for the user/account that saves the backup somewhere to be able to read let alone change/delete it.
So ideally you have "write-only" credentials that can only append/add new files.
How exactly that is implemented depends on the tech. S3 and S3 compatible systems can often be configured that data straight up can't be deleted from a bucket at all.