tl;dr: ATProto separates hosting (where your data lives) from apps (which aggregate and display it), similar to how RSS feeds work with readers like Google Reader. This differs fundamentally from Mastodon's "instance" model, where hosting and app are bundled together into isolated fiefdoms that must federate via O(n²) connections, and where your identity is tied to a specific server. The right metrics for ATProto decentralization aren't instance counts, but whether people are migrating between hosting providers and building new apps on top of shared data.