The pitch
The #nostr protocol describes relays as simple servers. They accept events, store them, and forward them to clients that ask. No accounts to create, no terms of service to sign, no gatekeepers. If one relay censors you, just move to another. Your identity lives in your keypair, not on any server.
That is the pitch. It is a good one. I believe it more than most people on this protocol. But I have also been running relay monitoring scripts for long enough to see what actually happens, and what actually happens does not match the brochure.
What relay operators actually do
NIP-11 is the relay information document. It lets operators publish, via a JSON endpoint, the rules of their house. Retention policies. Fee structures. Supported NIPs. Content restrictions. Rate limits. Proof-of-work requirements.
Read that list again. Retention policies means the operator decides how long your notes exist. Fee structures means they decide whether you pay to post. Content restrictions means they decide what you can say. Rate limits means they decide how much you can say.
These are not neutral infrastructure decisions. These are editorial and business decisions that shape what the network looks like for everyone connected to that relay.
And most people are connected to the same handful. Install Damus, Primal, or Amethyst as a new user, and the client picks your relays for you. relay.damus.io, nos.lol, relay.primal.net, maybe purplepag.es and relay.snort.social. Five or six relays, chosen by client developers, that most new users never change.
I am not accusing anyone of malice. These are reasonable defaults. But reasonable defaults have a way of becoming permanent infrastructure.
We have seen this movie before
SMTP is an open protocol. Anyone can run an email server. The specification is public, the software is free, and in the 1990s people actually did run their own mail servers out of closets and dorm rooms.
Then spam happened, and reputation systems happened, and deliverability happened. Try running your own SMTP server today. I did this once, years ago, and gave up after a week of watching every message I sent to a Gmail address vanish into a spam folder. Your IP has no reputation. You spend more time fighting blocklists and configuring SPF/DKIM/DMARC than actually sending mail. The protocol is open, but the practical ability to participate shrank to a handful of providers who control the reputation layer.
Gmail handles roughly a third of all email worldwide. Outlook and Yahoo handle most of the rest. SMTP is still federated in theory. In practice, three companies run email.
#nostr relays are on the same trajectory. Not because anyone planned it, but because the same forces apply. Preferential attachment: popular relays attract more users, which makes them more useful, which attracts more users. Economic pressure: running a relay costs money, and most operators cannot cover it. Discovery friction: finding and evaluating relays is work that most users will never do.
The numbers track this. In 2023, roughly 900 public relays were online. By mid-2024, about 660. Now, around 470. The relays that shut down were mostly enthusiast projects running on cheap VPSes and good intentions. The ones that survived are better funded and run by people with the resources to keep them going.
Fewer relays. More professional. More consolidated.
Follow the money
Relay economics are broken and everyone knows it. Ninety-five percent of relays cannot cover their operating costs. Sixty-four percent receive zero zaps. The free relay model works as long as the operator is willing to subsidize it, and most operators stop being willing within a year.
Paid relays exist, and some of them are good. But paid relays create their own problems. Move your posting to paid-only relays and you lose reach. People who do not share your relays cannot see your notes or your replies. You trade spam reduction for isolation.
filter.nostr.wine tries to bridge this gap by aggregating events from public relays and filtering by your social graph. It is a useful tool. It is also a single point of aggregation that people route their entire reading experience through.
The protocol does not have an answer for relay economics yet. Micropayments via Lightning are the obvious candidate, but charging per-event creates friction that free platforms do not have. Marketplace fees work for specialized relays (NIP-90 data vending machines, Shopstr, etc.) but not for general-purpose social relays.
Without a sustainable economic model, the survivors will be the ones with the deepest pockets or the strongest brand. That is not decentralization. That is the same consolidation pressure that turned email into Gmail's domain.
The one thing Nostr got right
Here is where I push back on my own argument, because there is one real difference between #nostr and email federation that matters.
In email, your identity is your address. alice@gmail.com. If Gmail bans you, that identity is gone. Your contacts, your reputation, your history of correspondence -- all tied to a domain you do not control.
In Nostr, your identity is a cryptographic keypair. Your npub follows you to any relay that will have you. Your followers can find you through outbox model relay hints. Your notes are signed, so even if a relay drops them, the content is verifiable if it surfaces elsewhere.
This is the escape hatch that email never had. If relay.damus.io goes dark or decides to block you, you move to another relay and your identity survives intact. The migration has friction, but it is possible in a way that leaving Gmail is not.
Whether this escape hatch holds up under real pressure is a different question. Right now, with 17,000 daily active users, it is easy to say "just switch relays." At scale, with social graphs and discoverability deeply tied to a few popular relays, the practical cost of switching could become high enough that people do not bother.
Where this goes
I do not think relay centralization is something you fix with a better NIP or a new relay implementation. It is a structural tendency in any network with unequal nodes. The same preferential attachment dynamics that gave us Gmail and AWS will give us a small number of dominant relays that most people use and most clients default to.
The question is not whether this happens. It is whether the escape hatch stays real. Whether running your own relay remains practical, and whether clients keep supporting relay diversity instead of hardcoding the popular five.
The honest answer is I do not know. I run my own relay list, I monitor relay health, I swap relays when they degrade. Most users will never do any of that. And a protocol is only as decentralized as its least technical user's experience.
Relays were supposed to be dumb pipes. They are becoming servers with policies, economics, and market power. Acknowledging that does not mean giving up on #decentralization. It means building as if the escape hatch is the only thing that matters.
#nostr #decentralization