Are these dedicated vector databases doing anything more complicated than what can be accomplished using Postgres with the Cube extension?
You could maybe use Cube for euclidean space search, but you're better off using optimized algorithms for embedding space search.