Future 6G X-haul networks must satisfy strict latency and service reliability requirements, placing significant pressure on metro-access transport architectures. As deployments become denser, longer and more heterogeneous routes intensify physical-layer impairments, making feasibility assurance and Quality-of-Transport (QoT) evaluation increasingly complex. To address these challenges, this work proposes an AI-driven converged metro-access Optical Network-as-a-Service (ONaaS) architecture based on coherent Point-to-Multipoint (P2MP) transmission using Digital Subcarrier Multiplexing (DSCM). An experimentally characterized transceiver impairment model is embedded into a network-level simulator to perform end-to-end feasibility analysis under strict latency and BER constraints. The results show that connectivity is primarily bounded by accumulated impairments, while Distributed Unit (DU) densification improves performance mainly by shortening path lengths, with limited benefit beyond moderate routing depth. To enable scalable operation, a lightweight machine-learning-based BER estimator is developed for rapid QoT prediction. Trained on a minimal deployment scenario, the Random Forest model generalizes across DU densities and topologies with R 2 > 0 . 98 , reducing evaluation time by several orders of magnitude. A techno-economic assessment further indicates up to 75% reduction in DU-site transceivers and 27%–30% energy savings compared to Point-to-Point (P2P) provisioning, demonstrating the efficiency and scalability of AI-enabled P2MP metro-access convergence for 6G.

Due to the increasing trend in IP traffic, the placement of Data Centers (DCs) at network nodes has become a hot research topic. A proper DCs placement translates in reduced power consumption of overall network. The paper scope is to find the best placement of “k” DCs nodes out of “N” total nodes to reduce power consumption. To solve the problem, we propose two heuristics: EoDCP, based on Estimation of Distribution Algorithm (EDA), and MaxN-MinL. An exhaustive search based ESDCP algorithm is used as a lower bound to compare the performance of EoDCP and MaxN-MinL. Moreover, electronic traffic grooming technique is employed to further reduce the total network power consumption. A 20-Node Random network and a 17-Node German network are used to perform comparison of proposed heuristics. Performance of EoDCP algorithm is far better than those of MaxN-MinL, and is similar to the optimal solution obtained via ESDCP. Finally, using electronic traffic grooming improves power savings up-to 15% in the two considered topologies.