Part of the point, as I understand it, is that moderation scales more-or-less linearly because each instance is ideally human-scale (that is, maybe having hundreds of members as opposed to millions) and responsible for its own moderation, unlike sites like Twitter or Facebook which have to pay a relatively small team of moderators to moderate millions of people's communications.

There's nothing saying any particular instance will moderate, but if they don't and it becomes a problem they'll quickly find themselves cut off from federating with other instances.