Based on the new threats posed by Artificial Intelligence(AI) to electromagnetic spectrum security and the current development status and trends of anti-jamming satellite communication systems, an intelligent anti-jamming satellite communication system architecture based on AI is proposed, which features 'distributed perception+hierarchical decision-making+unified action+quasi-real-time assessment'. The corresponding unit composition architecture of the satellite payload, terminals, and ground stations, and the hardware and software architecture required for engineering implementation are introduced as well in order to adapt to the increasingly intelligent interference situation. In light of the characteristics of extensive use of deep neural networks in the proposed intelligent anti-jamming satellite communication system architecture, the working mechanism of the system architecture is studied, the basic workflow and preliminary deployment methods are put forward. The relevant supporting technologies are analyzed and sorted out, the disadvantages of traditional technologies in areas such as interference perception, anti-jamming decision-making, anti-jamming waveform reconstruction, and quasi-real-time anti-jamming effectiveness assessment, are discussed. The current research status based on AI technology and the key issues that need to be focused on and resolved in the future are concluded. This is intended to provide some reference for the construction of AI-based anti-jamming satellite communication systems.