Skip to main content
bex.co
419
Blog

The Social Graph Nobody Owns: How Bluesky and the AT Protocol Became Web3's Default Open Network

9 min readDora NodaDora Noda
Social Media
Web3
Decentralization
identity
Farcaster
Share

Here is the uncomfortable fact that should keep every Web3 social founder awake at night: the most successful decentralized social network of this cycle did not launch a token, did not run an airdrop, and barely mentions the word "crypto." Bluesky crossed 43.5 million registered users by April 2026 — up from 10 million in September 2024 — by betting on something the token-incentive playbook keeps getting wrong. People do not want to own a social network. They want to never be locked out of one again.

That bet is built on the AT Protocol, an open, federated foundation for social applications. And in a quiet reversal of how the last decade was supposed to go, crypto builders are no longer trying to convince mainstream users to come to Web3's social graph. They are showing up to build on the one that already has the users.

The Backstory: How a Closed API Created an Open Protocol​

To understand why AT Protocol matters, rewind to February 2023, when X (then Twitter) shut off free access to its API. Overnight, an entire ecosystem of third-party clients — Tweetbot, Twitterrific, and dozens of analytics tools — went dark. The replacement tiers started around $100/month for hobbyist access and climbed into five figures for anything resembling research-grade data.

The damage radiated beyond nostalgic app developers. A generation of "InfoFi" projects — tools like Kaito that turned public social signal into market intelligence — lost their raw material in a single policy change. The lesson landed hard across both the social and crypto worlds: building a business on someone else's API is building on rented land, and the landlord can change the locks whenever the quarterly numbers demand it.

The AT Protocol is, in a real sense, the institutional memory of that trauma made into architecture. Its entire design answers one question: what would a social network look like if no single company could ever pull the plug on you?

Inside the Architecture: Identity That Outlives the Company​

Most "decentralized social" pitches collapse the moment you ask how identity works. AT Protocol's answer is its most important contribution, and it is worth understanding the pieces.

When you create an account, you get a Decentralized Identifier (DID) — something like did:plc:123abc — a stable cryptographic identity that belongs to you, not to any app. Your human-readable handle (a domain name) maps to that DID, but the DID is the real you. Switch apps, switch hosting providers, even leave Bluesky entirely, and your identity travels with you.

That identity points to a Personal Data Server (PDS), the source of truth for your "repository" — your posts, likes, and follows. Critically, the PDS is portable. Every record lives in a user-owned Merkle Search Tree, content-addressed by cryptographic hash, which makes the whole repo tamper-evident and trivially replicable. You can export your entire history and identity keypair and move to a different PDS without losing a single follower.

Sitting above the PDS layer are two more services:

  • Relays crawl thousands of PDS instances and aggregate their changes into a single network-wide "firehose" — a real-time stream of everything happening across the network.
  • AppViews consume that firehose and build the actual application experience: timelines, threads, notifications, search.

Holding it all together is Lexicon, a shared schema system. Instead of every app inventing its own incompatible data format, Lexicon defines common vocabularies (app.bsky.feed.post is a post, everywhere), so independent implementations interoperate by default. This is the part Web3 social networks chronically underbuild: a follow on one app is a follow on every app, with no bridge contract required.

The design goal has a name the protocol's engineers use deliberately: credible exit. The system counts as decentralized not because everything is already distributed, but because if Bluesky's company vanished tomorrow, every piece — your identity, your data, your social graph — could be picked up and operated by someone else. Ownership is defined as the guaranteed ability to leave.

The Honest Caveat: "Decentralization Theater"​

It would be dishonest to write about AT Protocol without addressing the criticism head-on, because it is substantial and it is mostly correct.

As of early 2026, roughly 99% of users are hosted on Bluesky PBC's own infrastructure. There is, in practice, one dominant relay (a machine that by late 2024 already needed around 5 terabytes of storage and is growing fast) and effectively one main AppView. An influential October 2025 assessment called the whole arrangement "decentralization theater" — the architecture permits decentralization, but the network operates as a centralized service with an escape hatch nobody has needed to use.

The fair response is twofold. First, "credible exit you haven't exercised" is still categorically different from "no exit exists" — which is exactly the X API situation that started this story. Second, the theater is slowly getting real actors. Blacksky now runs its own relay, PDS, moderation stack, and is building an independent AppView — the first genuine "second full provider" the network has had. In September 2025, Bluesky PBC announced it would hand the critical PLC identity directory to an independent Swiss association, removing one of the most centralized single points of control.

None of this makes Bluesky maximally decentralized today. It makes it plausibly decentralizing — which, for 43 million users who mostly just want a usable app, has turned out to be the more durable strategy than starting maximally decentralized and hoping users tolerate the friction.

Why Crypto Builders Stopped Competing and Started Building​

Here is where the Web3 angle gets genuinely interesting. For years, the decentralized-social thesis was a race: Farcaster, Lens, and others competed to build the crypto-native social graph that would peel users away from incumbents. The scoreboard tells a humbling story.

  • Farcaster placed identity onchain and built a polished product, but hit scaling walls and ultimately leaned toward wallet and app infrastructure. Registering still costs users real crypto (historically around $5/year), a small but real onboarding tax.
  • Lens Protocol went fully onchain with profile NFTs and publications on its own L2 stack — maximal composability, but with the throughput and UX friction that onchain-everything implies.
  • Bluesky skipped the blockchain entirely, used boring federation, and won the user numbers by an order of magnitude.

The realization reshaping the space in 2026 is that these are not necessarily competitors. AT Protocol is, functionally, a public, permissionless social data layer — an open firehose of human activity that anyone can read and write to without asking Bluesky's permission. For builders who learned the X-API lesson, that is the prize. You do not need to bootstrap your own 40-million-user graph if you can build a specialized app on top of one that exists, inheriting its follows, feeds, and identities through Lexicon.

The emerging ecosystem — the "ATmosphere" — already shows the shape of it. There is Tangled for git-style collaboration, Leaflet for long-form publishing, and in May 2026 Automattic shipped an ATmosphere plugin that pipes WordPress content directly into the network as native records. None of these are "Bluesky features." They are independent applications sharing one identity and one social graph — the composability Web3 social always promised, delivered by a protocol that mostly refused to call itself Web3.

What This Means for the Next Five Years​

The convergence is the story. Crypto-native projects bring what AT Protocol still lacks — value transfer, verifiable scarcity, token-curated incentives, and onchain settlement — while the AT Protocol brings what crypto social never achieved: tens of millions of real users and a portable identity layer that mainstream people actually adopted by accident.

Expect the integration to deepen along three axes. Identity bridges that let a did:plc identity carry an onchain wallet (and vice versa) turn a Bluesky handle into a payable, ownable address. Open social analytics rebuild the InfoFi tooling that X's API shutdown destroyed, this time on a firehose nobody can switch off — Kaito-style attention markets, but on infrastructure with credible exit baked in. And specialized clients monetize through crypto rails (tips, subscriptions, token-gated feeds) on top of a graph they did not have to build.

The deepest lesson cuts against a decade of Web3 social orthodoxy. The winning move was not "put the social graph on a blockchain." It was "make the social graph a public good that no company can repossess, and let the blockchain handle the parts blockchains are actually good at." Ownership turned out to mean the right to exit, not a token in your wallet — and that reframing is going to outlast a lot of the projects that bet the other way.

For developers, the practical takeaway is concrete: the social layer is becoming open infrastructure, the same way RPC endpoints, indexers, and data availability already did. The teams that win the next cycle of consumer crypto will be the ones who treat identity, social graph, and value transfer as composable services to assemble — not moats to rebuild from scratch.

Building applications that bridge open social graphs and onchain value? bex.co provides reliable RPC and indexing infrastructure across the chains where identity and payments settle — so you can focus on the product, not the plumbing.

Sources​

Related articles

Learn with a bex course

A short, practical course — from primitives to production — built from the bex docs and the corpus you just read.

Start course
Back to all posts