1. Sam makes a post.
2. FB predicts Pat will comment on this post if they see it.
3. Pat sees the post, and writes a comment.
4. FB predicts Alex will reply to Pat's comment if they see it.
5. Alex sees the comment, and writes a reply.
To show why data transfer is such an issue, assume Sam, Pat, and Alex are all in jurisdictions with EU-style privacy regulations and that don't have data transfer agreements with each other.
How would you build a system that supports 1-5, a user journey that is core to Facebook's usage, in a way compliant with these regulations? For example, where is the discussion stored? Where do the models in (2) and (4) run?
As far as I can tell, the really difficult aspect here is how and where to permanently store the fact that the two users are talking to each other once the comments are actually made, since the mere fact that they are talking to each other demonstrates a relationship between them which may be considered PII in some contexts. Or at least, it would be difficult if the US also had privacy laws like the EU's, and IMO any coherent solution should be able to work if the US adopts something like GDPR. Unlike the message contents, this is quite difficult to store in a privacy-preserving way. I think the discussion would be more interesting and feel less like attacking a strawman if people were to focus on the interesting questions like this one, rather than the extremely uninteresting question of whether Facebook can serve posts to the US at all (which it obviously can).
For those downvoting me, I'd love to hear why you think it's more complicated than that. Contrary to popular belief, the people who wrote the GDPR are pretty technically sophisticated and understand that data on the internet has to move between countries from time to time.