This requires the whole plaintext/cipher text to fit into memory. For small blobs I would agree that GCM is perfect, but for anything larger you end up with extra projects like https://github.com/minio/sio trying to break streams into smaller ordered chunks. They are on v2 now.
