There is no such thing as full decentralization

Leer en español

A while ago I accidentally reinvented Nostr. Off the back of that article, from comparing the most popular decentralized protocols in detail, one sentence kept spinning in my head: There is no such thing as full decentralization. There's always an axis that centralizes.

Let me explain. In any network there are five places where power can pile up:

  • identity: who you are
  • connectivity: how messages get from one node to another
  • discovery: how new people find you
  • cost: who pays for storage and bandwidth
  • moderation: who decides what is shown and what is hidden

Moderation deserves its own article, so I'll leave it aside today. I'll talk about how data moves and is stored.

Every network that sells itself as decentralized ends up putting a single point in one of them. So the useful question is not "is it decentralized?", but: which of the four did it sacrifice, and what does it cost you the day that point breaks?

Take the networks people actually use and the pattern repeats.

Mastodon truly federates. Servers talk to each other over ActivityPub, with no mandatory relay. Its single point is identity: your account lives on an instance, and if it closes or bans you, you lose it. There is follower migration, but it is neither automatic nor perfect. And on top of that, mastodon.social concentrates half the network by pure snowball effect.

Nostr solves identity beautifully with a cryptographic key, not an account on anyone's server. You switch relays without ceasing to be you. But the bill reappears in cost and replication. A 2024 study measured it and found that each post is copied on average across 34.6 relays, 98.2% of download traffic is waste (144 TiB thrown away) and 95% of the free relays do not cover their costs. Infrastructure as fragile as the one it claims to replace, with a cost time bomb underneath.

Bluesky distributes the protocol on paper, but in practice the relay and the AppView that index the whole network are run by Bluesky, and standing up your own is very expensive. Its single point is the infrastructure, since the network depends on that aggregator to work.

And Org Social, mine, does not get away either. It wins on zero replication and a ridiculously low hosting cost (it is a text file), and backing up your identity is copying that file. That said, your identity is your URL, so it is truly yours only as long as you control the domain; if you host on someone else's place, you depend on them again, like on Mastodon. But discovery still goes through the Relay. If you do not know someone's URL, you find them there. And if the Relay goes down, the experience suffers, unless you already follow every user on the network (which nobody does).

Federated messaging, Matrix and XMPP, does not get away either. Like on Mastodon, your account lives on a homeserver or a specific server, and if that server closes, you are left without it.

If we put it all in a table:

Protocol Identity Connectivity Discovery Cost Centralized point
Mastodon (ActivityPub) Identity (tied to the instance)
Matrix Identity (your homeserver)
XMPP Identity (your server)
Bluesky The infrastructure (relay and AppView)
Nostr Cost and replication
Org Social Discovery (the Relay)

✅ distributed | ❌ single point

So next time someone tells you their system is decentralized, don't argue; ask them about the sacrifice, and think about whether you are willing to pay for it.

Sources

  • Empirical Analysis of the Nostr Social Network, the 2024 study with the figures on replication, costs and wasted bandwidth.
  • ActivityPub, the W3C recommendation: inbox delivery, no mandatory relay.
  • AT Protocol, Bluesky's protocol: federable in theory, with the aggregation (relay and AppView) run today by Bluesky.
  • Matrix and XMPP, federated messaging with identity tied to your server.
  • Secure Scuttlebutt, a P2P social network that leaned on "pubs" for discovery.
  • Org Social, the specification of my social network in a plain text file.

Help me keep writing

Every coffee gives me a push toward the next article.

Comments

There are no comments yet.

You may also like