AI/ML growth is driving a surge in interconnect bandwidth as models and datasets scale and workloads become increasingly token-centric—reshaping system design. Modern systems have converged on a two-tier hierarchy: scale-up fabrics that tightly couple XPU, memory, and switching into a coherent “virtual XPU,” and scale-out networks that connect many scale-up fabrics across the data centre. Interconnect technology choices are constrained by power, bandwidth density, latency, reliability, and cost. This symposium will explore the evolution of scale-up and scale-out interconnects—which technologies are most likely to win, why, and what key attributes will be required for next-generation AI/ML workloads.
The use of AI in society is exploding and quickly changing how businesses and public functions operate. It’s speculated that Generative AI could drive a 10x traffic growth in 6G mobile networks. This workshop will outline the background of such predictions and will discuss possible 6G architectures and technologies to cope with such a massive traffic growth. This will lead to the needed optical technologies for the 6G AI era with the question: can optics solutions for a 10x traffic growth can be realised with reasonable cost, size and power consumption and can optical standardisation meet these needs?
AI workloads, IMT-2030 (6G) transport and hyperscale data-centre optics are converging, forcing networks toward 800G/1.6T interfaces, microsecond-level synchronization, deterministic latency, and stringent energy budgets. ITU-T SG15’s ION-2030 technical report is the first cross-domain blueprint unifying access, metro/core and DC fabrics. This symposium aims to translate ION-2030 into deployable, standards-aligned architectures—AI-native operations (agents, digital twins), converged DCN/DCI/DCA, VHSP PON and ISAC—delivering predictive, SLA-assured and autonomous NaaS across E2E optical networks, with enhanced security, resilience and integrated sensing—providing interoperable roadmaps, measurable KPIs and practical guidance for operators, hyperscalers, vendors and researchers.
Programmable Integrated Photonics (PIP) represents a paradigm shift from application-specific PICs to reconfigurable optical subsystems integrating hardware, electronics, and control algorithms. Recent advances in scalable integrated photonics and self-configuring waveguide meshes/multi-port interferometers make this approach gradually ready for realistic deployments that address the growing needs and challenging requirements of existing and emerging applications that appear at an accelerated rate in the era of AI. As the industry seeks adaptable optical subsystems with shorter development cycles and multi-functionality, the symposium aims to address a wide gamut of emerging applications spanning communications, sensing and computing domains.