When I do video-editing, its rare for my video editor to go above 10 cores (20 threads). Only Blender3d really benefits from such huge core-counts, and even then, GPU-rendering is one of the fastest (and cheapest) ways to improve that experience. The number of desktop applications that even benefit from 16 cores, let alone 32 or 64, is miniscule.
Adobe Premier? No. Blender? Only for the Cycles renderer and none of the plugins actually speed up mind you (single-threaded Python warning). Large scale compiles for programming? A little bit, but not to 32 or 64 level. CMake/Ninja really helps with parallelization, but so does partial builds in practice. Deep Learning? GPU-based these days.
64-core workstations seem like they're for the exact subset of Walt Disney 3d modelers. (100GB+ scenes that can't fit on GPUs and must be CPU-rendered). Sure, a niche but an ever shrinking one.