We're actually well beyond RAID arrays. Google CEPH. It's actually both super complicated and kind of simple to grow to really large storage amounts with LOTS of redundancy. It's trickier for global scale redundancy, I think you'd need multiple clusters using something else to sync them.

I also always come back to some of the stuff freenet used to do in older versions where everyone who was a client also contributed disk space that was opaque to them, but kept a copy of what you went and looked at, and what you relayed via it for others. The more people looking at content, the more copies you ended up with in the system, and it would only lose data if no one was interested in it for some period of time.