Abstract: The ever increasing scales of AI applications mandate the exploitation of multiprocessor-based parallel AI processing systems. This paper presents a concise outline of the multiprocessor architecture features in support of AI, addresses the general idea of parallel AI processing, and emphasizes the importance, for efficient parallel AI solving, of offering multi-level and multi-dimensional support, special hardware support, flexible structural support, architecture oriented programming support, etc. Finally some major project or product examples are given.