Maybe, except from my understanding an embedding vector tends to be much larger than the source token (due to the high dimensionality of the embedding space). So it's almost like a reverse compression in a way. That said I know vector DBs have much more efficient ways of storing those vector embedding.