Cambricon Technologies — AI supply-chain exposure

Cambricon designs AI accelerators — its Siyuan/MLU line of ASICs built to train and run neural networks — but like most chip designers it owns no fabrication of its own. It sits in the logic layer of the chain, drawing the blueprints for silicon that someone else must manufacture and package. Within China it has become the listed pure-play name investors reach for when they want exposure to the country's push to substitute domestic accelerators for imported GPUs.

Its structural hook is less any single chip than what it represents: a policy tailwind. As export controls tighten the supply of foreign accelerators into China, demand concentrates on the handful of domestic designers able to offer a credible alternative, and Cambricon is among the most visible. The model places it centrally on that substitution thesis — its relevance rises or falls with how hard the domestic-silicon mandate is pushed, not with the merits of any one part number.