Abstract:The rapid advancement of space missions has created an urgent demand for real-time intelligent processing capabilities in orbit within the aerospace sector. Establishing a highly intelligent computing architecture represents an effective solution to meet these in-orbit real-time processing requirements, with spaceborne AI chips serving as the key enabler. Faced with increasingly complex space environments and computational demands, spaceborne AI chips developed based on traditional closed-source instruction set architectures face challenges such as licensing risks and energy efficiency limitations. Early hardware architectures also exhibit insufficient computational power when handling complex AI inference tasks. Consequently, exploring new architectural paradigms and application models has become an industry consensus for spaceborne AI chips. This paper reviews the current application status of spaceborne AI chips globally, systematically analyzes the development status and trends of the industry, and summarizes key trends in areas such as open-source instruction set architectures, heterogeneous hardware architectures, and multi-satellite multi-intelligence systems. These insights aim to provide a reference for relevant technological R&D and industrial planning.