LLMs' coding prowess hasn't helped ROCm or Intel stacks catch CUDA

A Reddit post questions why NVIDIA's CUDA remains dominant despite LLMs' coding advances, pointing to slow progress in AMD's ROCm and Intel's GPU software ecosystems. The author argues that without competitive alternatives, NVIDIA can charge a premium for its "it just works" stack.