A cyber digital twin is a bridge connecting cyber and physical space, and it also provides a paradigm for intelligent service-oriented future network architectures. This article proposes a cyber digital twin-based future network architecture to support cross-space intelligent services and manage cross-space resources to meet the demand for diversified intelligent services. The cyber digital twin is introduced as a smart communication agent in the edge network to handle all physical object communication; thus, all the physical object data is acquired by its cyber digital twin. To meet the demand for diversified intelligent services, the proposed network architecture consists of a service analytical model, cross-space resource representation and management, and an autonomous configuration model. The service analytical model realizes the intelligent matching of new service requirements to resource requirements. Furthermore, the cross-space resource representation and management mechanism is based on reinforcement learning to ensure cross-space service consistency and resource utilization efficiency. Finally, the autonomous configuration model provides a unified intelligent configuration method for a large number of complex operations with cross-space devices. This article also proposes a cross-space resource management scheme that utilizes reinforcement learning to learn the optimal resource allocation strategy. The simulation results demonstrate its excellent performance.