I'm not happy that it introduces yet another data object library, but at least it's possible to build plugins for existing ones -- SQLAlchemy and Django are already there, and it would be nice to add ones for attrs, Schematics and of course the 3.7 native data classes.
Regarding the popularity of REST -- even if it wasn't intended for CRUD originally, it fits perfectly with that paradigm, and I think this is why many people were able to implement it easily when they started using it.
Moreover, the community around it in both its GitHub and StackOverflow isn't as active as the equivalent GraphQL communities in other languages/ecosystems.
All in all I believe that increased adoption of GraphQL will lead to even better tooling and I'm sincerely hoping it'll catch on :).
My primary 'beef' with REST has always been the "10 ways to do 1 thing" especially when it comes to less standardised features like pagination, type-definition/validation, etc. GraphQL introducing some of these things out-of-the-box would hopefully reduce the comfort-zone-based bikeshedding that comes with any new REST API being developed.
I quickly learned why Django rest framework is the defacto standard... The tooling is just uncomparibly better
I haven't seen anything that would make me advise GraphQL over REST, specially now that REST tooling is finally catching up with validation and graphical generation.
All the presentations I have seen about how productive GraphQL is, seem to live in JavaScript with NoSQL databases silos.
Working in Python-land I'm pretty confident REST wouldn't have been as ubiquitous if it wasn't for tools like Flask, marshmallow, and the myriad of other tools.
Having taken up GraphQL in personal projects I can say that at least in Python the tooling just ain't there so until it gets there I don't see GraphQL killing anything.