Friday, January 23, 2026
CUDA to ROCm Porting
People are discussing how Claude Code rapidly ported CUDA code to ROCm, potentially disrupting the semiconductor industry's power dynamics and challenging NVIDIA's CUDA dominance.
AMD just got a massive bullish catalyst — time to consider adding to AMD positions! Core of the event: On January 22, 2026, a user named ohnnytshi used Claude Code and in just 30 minutes successfully ported a CUDA backend to ROCm. And he had never written a kernel before in his
Claude Codeがわずか30分でCUDAをROCmへ移植:崩れ去る「CUDAの掘」とAIが書き換える半導体業界の勢力図 https://xenospectrum.com/claude-code-ports-cuda-to-rocm-nvidia-moat-analysis/… いよいよか、、、 [Links: https://t.co/72oDdwpG9M]
I feel it is indeed a lot easier to write a backend for every operators on ROCm with claude code so CUDA indeed has no moat in inference. However, they have way bigger moats somewhere else.