with a proper algorithm which may well not exist it's in theory possible to e.g. perform deep learning on all training examples in a single pass or something like that.
It's a bit like saying that QC will be awesome because thanks to it people could in theory find the cure for cancer, in some way, somehow.