> put on AI inference when they don't build the GPU?

GPU in AI world is essentially a set of specific matrix calculations that this CPU supports on hardware-level. Thought memory speed seems low.