As CSPs look for new sources of growth, one opportunity is the integration of computing and networking services to meet the growing demand for AI. However, the hardware and electricity needs for model training and inference work are increasing exponentially, making sustainable electricity a huge challenge and even a bottleneck for AI development. The distributed deployment of computing resources, intelligent scheduling of computing power, and efficient use of the network itself could help address this issue. To that end, this Catalyst encompasses two streams. The first stream will harness digital twins and AI to reduce the computational costs of AI training and inference and enable low-energy computing network services. Digital twins can provide a comprehensive and real-time view of computing power and network resources, including their energy status. AI models can then apply intelligent decision-making, simulation, and trial-and-error, ultimately outputting configuration parameters to physical devices. By optimizing paths and dynamic scheduling across regional data centres, this approach can improve the efficiency of computing power and network resources employed to train large models. The second stream is focused on providing one-stop computing network services for SMEs. The goal is to enable CSPs to combine multi-domain network products (F5G-A, 5G, Wi-Fi 7, IoT), public clouds, specialized platforms, AI models and generative AI live broadcast applications to provide full-stack services for SMEs. Using cloud dedicated lines and inter-cloud connections, based on the latest OTN high-speed lossless connections and SRv6 technologies, the solution will interconnect various heterogeneous resources and perform intelligent orchestration and scheduling based on business intent. Blockchain and multi-dimensional billing technologies will facilitate transactions among partners. The overall architecture remains consistent with that in the first phase of the Catalyst, which employed TMF Open Digital Architecture components and APIs. The second phase also plans to employ the TMF Open Gateway interface (TMF931) and intends to contribute an Operate API. The overarching goal is to achieve industry-leading energy consumption and network resource utilization in demanding scenarios, such as large model training and video rendering, while generating new revenues for CSPs and their customers.