And I am guessing ML workloads use up a single computer to do just that (and nothing else), meaning that NVidia could start their own fork of Linux, and people need to just eat it (or switch to AMD, Intel, etc.!)