I wish these guys luck, but I'm becoming increasingly pessimistic about the idea of a 100% edge-only decentralized network that is really robust and useful.

I don't think it's just a matter of putting the engineering effort to bear. I think there are fundamental mathematically-based barriers here. Try this paper for starters:

https://www.zerotier.com/misc/2011__A_Little_Centralization_...

The CAP theorem is also very relevant.