The exponential growth in AI computing demand is pushing the limits of traditional networking architectures. While per-GPU computational power continues to advance, physical and thermal constraints increasingly challenge the scalability, reliability, cost, and efficiency of current interconnect solutions. Optical networking, particularly through Optical Circuit Switching (OCS), is emerging as a promising approach to address these challenges, enabling scalable, reconfigurable, and energy-efficient infrastructure for AI clusters.
This workshop will examine both the application-driven requirements and the enabling technologies for OCS in AI datacenters. It will bring together leading voices from academia and industry to share perspectives on system-level use cases, performance benchmarks, orchestration strategies, and technological advances in MEMS, LCoS, and silicon photonics. The session is designed to provoke debate not just on the technical feasibility, but also on the market adoption dynamics, competitive landscape, and future roadmap for OCS.
The key questions to address in this workshop are:
- Beyond one dominant early adopter, is there real market momentum for OCS in AI datacenters?
- What are the applications truly driving optical switching demand—does OCS go beyond functioning as a “dumb” optical patch panel?
- What unique benefits does OCS deliver compared to conventional switching approaches?
- What risks could delay or even derail market adoption? Is heterogeneous software integration the biggest barrier?
- Volume challenge: Will fragmentation—too many suppliers chasing limited demand—undermine scalability and production capabilities, or will production be too accelerated by the few established suppliers?
- How do conventional electronic switching networks and CPO-based networks fit into the roadmap? Are they delaying, complementing, or competing with OCS deployments?
- Could OCS become the backbone technology for exascale AI clusters within the next decade?
- Is there a risk that advances in electronics (e.g., co-packaged optics or photonic-electronic integration) could leapfrog OCS entirely?
- What role should standards bodies and open ecosystems play in ensuring interoperability and accelerating the adoption of OCS?
Session 2: Enabling Technologies for OCS: This session will highlight and compare various OCS technologies such as MEMS, LCoS, and silicon photonics (SiPh), featuring industry leaders sharing their latest developments.
Organizers
-
George Michelogiannakis
Lawrence Berkeley National Laboratory, United States
-
Osamu Moriwaki
NTT, Japan
-
Daniel Pérez-López
iPRONICS Programmable Photonics S.L., Spain
-
Yvan Pointurier
Huawei Technologies, France
Speakers
-
Kazuya Anazawa
NTT, Japan
Kazuya Anazawa is a researcher at Network Innovation Laboratories, NTT Corporation, Japan. He received B.E. and M.E. degrees in computer science and engineering from the University of Aizu in 2016 and 2018. Since joining NTT in 2018, he has been involved in research and technological development of open optical transport systems and optical-circuit-switching systems for data centers and high performance computing clusters. He serves as a coordinator for reference implementation model for AI interconnect infrastructure working item in DCS-TF at IOWN Global Forum.
-
David Boertjes
Ciena, Canada
-
Peter Roorda
Lumentum, United States
-
Rachee Singh
Cornell University, United States
-
Luis Torrijos
iPronics, Spain
Dr. Luis Torrijos is Head of Photonics at iPronics, where he leads the development of advanced photonic integrated circuits for AI-driven data centers. His work focuses on creating large-scale silicon photonics optical circuit switches combining low latency, low cost, and low power consumption. He earned his PhD with honors from the Universitat Politècnica de València, followed by a two-year postdoctoral fellowship in programmable integrated photonics. He has contributed to the commercialization of the world’s first programmable photonic processor, authored papers in high-impact journals and international conferences and lead the developed of several patents.
-
Ming Wu
nEye Systems, United States