Abstract: We have now entered an era of ubiquitous intelligent connectivity, characterized by the integration of generative artificial intelligence and the fusion of social, cyber, and physical elements (socio-cyber-physical integration). The resulting intelligent systems and applications require novel basic software support. In terms of resource management, the concept of cloud computing needs to be extended to the ternary space of social, cyber, and physical resources. This involves achieving cloudification and on-demand provisioning of these resources through a “software-defined” approach. In terms of model inference, there is a need to integrate AI models of varying scales and domains for diverse scenarios and heterogeneous computing power, supporting hybrid deployment and differentiated inference quality-of-service (QoS) requirements. In terms of application frameworks, large model agents should form the foundation for the on-demand composition of social, cyber, and physical resources and related software services, as well as the trustworthy planning and execution of user tasks. This article elaborates on the relevant background, analyzes and discusses the technical architectures addressing the above three aspects, and prospects future development directions.