"NOTE: You may choose to validate assertions on your own server. While a bit more complicated you can reduce your dependencies on others. Refer to the specification and the source for the reference validator."
So while it's easier to use their server, it appears it is entirely possible to implement without it. Perhaps in a few weeks, we will see some good implementations. As I understand, the main advantage over OpenID is that instead of needing a <username>@<someopenidprovider> login, people can simply use an email address as their login. If all email providers supported OpenID, there would be little difference, but that has not happened (yet).
"NOTE: You may choose to validate assertions on your own server. While a bit more complicated you can reduce your dependencies on others. Refer to the specification and the source for the reference validator."
So while it's easier to use their server, it appears it is entirely possible to implement without it. Perhaps in a few weeks, we will see some good implementations. As I understand, the main advantage over OpenID is that instead of needing a <username>@<someopenidprovider> login, people can simply use an email address as their login. If all email providers supported OpenID, there would be little difference, but that has not happened (yet).