Advanced Micro Devices

Moat sources are multi-pronged but not indestructible. Intangible assets include x86 architecture expertise and a patent and cross-licensing base built over decades, plus a fast-improving AI software stack (ROCm) and developer cloud that lower switching friction from CUDA.

Switching costs exist in enterprise CPU platforms and embedded FPGAs, and are emerging in AI systems as hyperscalers validate MI300/MI350 at scale. Cost advantages come from design scale and close foundry partner alignment.

Efficient scale is present in certain niches like console semi-custom and HPC where AMD powers the top two supercomputers and 172 systems on the Top500 list. Risks to durability: Nvidia’s stronger network effects in CUDA, hyperscaler custom silicon, export controls, and reliance on TSMC for leading-edge nodes.

Net-net, we assess a narrow-to-strong-narrow moat that could widen if ROCm adoption continues and system-level offerings mature.