Workshop: Convergence of Scale-Out and Scale-Across Networks: Architecting Resilience for the AI Era

For decades, intra-data center networks ("Scale-Out") and WDM data center interconnects ("Scale-Across") followed distinct technological evolutionary paths. Scale-Out DCNs relied on short-reach, O-band links using Intensity Modulation and Direct Detection (IMDD). Conversely, Scale-Across DCI networks leveraged dense C-band WDM, coherent transmission, optical amplification (EDFA), and wavelength switching (WSS).

Could the explosive growth of massive AI workloads and distributed training clusters be pushing for a technological convergence? Scale-Out networks are adopting Optical Circuit Switching (OCS), optical amplification, and coherent transceivers—notably 1.6T "coherent lite"— alongside multi-wavelength schemes (e.g., FR8, multi-lambda coherent lite) to span growing campus distances and overcome higher link losses. Emerging Hollow-Core Fiber (HCF) could further disrupt this boundary by enabling practical WDM transmission in the O-band, where low chromatic dispersion and associated four-wave mixing (FWM) penalties make dense wavelength transmission challenging in conventional fiber.

Simultaneously, Scale-Across DCI networks are adopting point-to-point pluggable optics (e.g., ZR/ZR+) over shorter sub-100 km links, while facing strict low-latency and fault-tolerance demands traditionally confined to the data center fabric.

This workshop explores whether the boundaries between O-band/IMDD-Coherent Lite and C-band/Coherent-WDM are permanently dissolving. Beyond hardware integration—such as unified DSPs, shared switch ASICs, and integrated optics—speakers and attendees will debate:

• Architectural Convergence: Will a single, unified optical stack span both intra-datacenter and inter-datacenter links, or will physical layer constraints enforce permanent boundaries?
• Reliability & Latency: How do strict SLA metrics for large-scale distributed AI training (e.g., collective communication latency, rapid fault isolation) translate across hybrid DCN/DCI boundaries?
• Coordinated Resilience: Can unified control planes and joint recovery strategies mitigate optical line failures before they stall multi-thousand GPU jobs?

Join leading experts from hyperscalers, optical system vendors, transceiver manufacturers, and academia as we debate the future blueprint of resilient optical infrastructure for the AI era.