Drives indeed fail, but if you look at the usual annualized failure ratio figures, they don’t fail even remotely enough to care (for Storj node purposes), unless they are very old. It’s actually way more common to see failure of other, replaceable components of the storage system. I can live with a yearly risk of 2% of a drive failure. Right now I’m running nodes on five drives, all close to full. One of them fails, I’ve still got four more.
Besides, of these 2%, not even all failure modes actually matter for Storj. Failures where a small part of your drive develop bad blocks is pretty much nothing. One of the drives I run for Storj has badblocks for many years—it still gave revenue multiple times it was worth. Yet regular RAID would immediatelly mark your drive as bad and consider your array degraded.
RAIDs and stuff is great for when you cannot lose even a single byte of your dataset. Losing 2% of Storj dataset is considered acceptable.