Ask HN: Path to AI GPU Programming

Regarding AI, the thing I’ve heard the most lately is “during a gold rush sell shovels”.

It’s a great advice (?) for a semiconductor company ready to get into the gpu market and compete with Nvidia, but what about me? I’m just a machine learning engineer bored to death by data cleaning and writing over and over the same two lines of PyTorch. I guess in that metaphor I’m the one using such shovels.

Perhaps building shovels is much more interesting than using them? What if I’d want to jump into CUDA kernel development? Is that a niche hard enough that the competition is low? Say I’m willing to put in the time and effort, where should I start? Best resources on the topic?

FYI, I’m well versed in C++ but zero in GPU programming