Federated platforms are far easier to build, develop for, divvy up moderation responsibilities, and finance rather than “truly decentralized” platforms which I believe pulls in far more risk (how much fun is it running a Tor exit node?), harder to fund (where would Tor/etc be without large universities and charities propping it up?), more content moderation issues (gl;hf dealing with CSAM if you build a ‘completely uncensorable, decentralized’ platform), and so on. At least utility cryptos (e.g Namecoin as mentioned) help solve the finance/commerce issue for some ‘truly decentralized’ ideas.
I don’t think servers are so much the problem. The problem is we have such an adversarial internet backbone and core internet infrastructure now that’s actively trying to prevent routing around censorship intentionally.
Despite my aforementioned concerns with the sustainability of the Tor network, I believe it’s a fairer option (or perhaps I2P, or whatever other overlay networks come about) to be hosting services on Tor/etc instead of clearnet.
Speaking in context of onion services exclusively (and not about interop with clearnet):
- Domain seizures on clearnet? So what, nobody can ‘seize’ your onion address on Tor, unless they legitimately have your private key.
- ISP shuts you down? So what, move the server elsewhere, come back online, nothing with addressing or any configuration changes at all.
- Stuck behind CGNAT and can’t self-host? So what, connect to the Tor network, and you can start hosting services on Tor/etc regardless of what your network topology is. Now everyone can self-host and spread out more.
- Afraid of rogue CAs, Cloudflare, and other TLS MitM? So what, the Tor network provides encrypted tunneling that only terminates at the holder of the private key for the respective onion address, far simpler than involving a Certificate Authority or delegated trust system.
I’m sure a response could probably be “well we tried, but nobody really bothers using an onion counterpart fedi server”, and honestly that’s because: a lot of fedi server software legitimately sucks for high-latency networks, especially for things that are heavily client-side rendered. All we need is some simple fedi server implementations that lean more on server-side rendering, and then the experience is far less miserable than waiting a literal minute or so for a Misskey profile to load (for example)