The discussion around atproto and Bluesky often hinges on a false premise: that there are “instances” to count. In reality, atproto separates hosting from aggregation, so no discrete instances exist. This misinterpretation echoes the Mastodon model, where each instance bundles host and app into a single box and each user can choose where their data resides today.

Before atproto, the web relied on simple hosting and app pull models. Bloggers owned their content on personal sites or platforms, while readers used aggregators like RSS feeds, Google Reader, or Feedly to compile posts. Those apps never stored the posts; they merely displayed them, keeping the original hosting independent and allowing users to move content freely today.

Mastodon flipped that model by bundling a single server into a instance. Users signed up at a domain, and their identity became tied to that server, e.g. [email protected]. When users followed someone on another instance, the two servers negotiated a federation link, forwarding public posts across the network. This created isolated, country‑like communities that could refuse to interoperate.

Atproto removes the box entirely. Data lives on a user‑chosen host, while any number of front‑ends can read it via the same protocol. A user can hop hosts with a single action, and developers can build new apps without touching the core database. This model scales like RSS, allowing true decentralization without the siloed “instance” narrative today everywhere.