Unfortunately I think, at least for now, that's just the way it is. This is why I personally focus on hardware architectures / acceleration for machine learning and biologically plausible deep learning. Ideas tend to matter more than compute resources in these domains.