GPUs are more useful than CPUs for ML, but eventually someone's going to want to cut nVidia out of that process: https://cloud.google.com/blog/big-data/2017/05/an-in-depth-l...

nVidia is still a solid choice as VR has raised the visual fidelity demands for games and experiences, and even a 1080Ti can't render VR games at 90+FPS at max graphics. That's still a lot of demand nVidia has to meet.