The current assumption made by these companies is that AI training is fair use, and is therefore legal regardless of license. There are still many ongoing court cases over this, but one case was already resolved in favor or the fair use position.

If an instance is merely blocked, does that mean all content produced by that instance, or by a Lemmy.World user using that instance, is strictly not stored on Lemmy.World servers?

Otherwise there might still be liability. Also, in the US you don’t even have to do anything illegal to be the target of a lawsuit—distancing from piracy is a practical defense against the cost of legal proceedings, even if it’s technically legal.