The Open Compute Project Foundation (OCP) has launched a new Optical Circuit Switching (OCS) Subproject aimed at accelerating the adoption of photonic switching technologies in AI data centers. Co-led by iPronics and Lumentum, the subproject brings together industry players including Google, Microsoft, Coherent, Lumotive, nEye, Oriole Networks, and POLATIS (HUBER+SUHNER) to develop open, interoperable solutions for high-bandwidth, low-latency networking.
Unlike traditional electronic switches, OCS relies on photonic circuits to route data, offering improved energy efficiency and reduced latency. The project aims to deliver compact, fast-reconfigurable optical switching that supports AI clusters running generative AI and large language models. Google, a long-time adopter of OCS in its Jupiter/AI network architectures, will contribute by helping define standard APIs and software interfaces leveraging gNMI, gNOI, gNSI, and OpenConfig.
The initiative will make its public debut at the OCP APAC Summit in Taipei on August 5–6, in a joint track with the IOWN Global Forum. OCP and IOWN leaders emphasized the importance of ecosystem collaboration in making sustainable, scalable, and energy-efficient infrastructure a reality for hyperscale AI workloads.
- OCS Subproject launched under Open Compute Project Foundation
- Co-led by iPronics and Lumentum with support from Google, Microsoft, Coherent, Lumotive, and others
- Focus on photonic switching to improve AI cluster efficiency and scalability
- Google highlights role of OCS in its TPU-based networks and Project Apollo
- Integration with software frameworks such as gNMI, gNOI, gNSI, OpenConfig
- First public presentation scheduled for OCP APAC Summit, August 5–6 in Taipei
“The Open Compute Project offers a unique chance to democratize access and drive the development of open, scalable solutions that meet evolving market needs and shape the future of computing,” said Daniel Pérez-López, cofounder and CTO at iPronics.
🌐 Tech Explainer: What Is Optical Circuit Switching (OCS) and Why It Matters for AI Infrastructure
Optical Circuit Switching (OCS) is an emerging class of photonic networking technology designed to bypass traditional electrical packet switching in data centers. Rather than converting optical signals to electrical and back again, OCS routes data entirely in the optical domain using tunable or reconfigurable photonic paths. This results in significantly lower latency, higher throughput, reduced power consumption, and better scalability—key attributes for next-generation AI and high-performance computing (HPC) clusters.
OCS vs. Traditional Switching:
- Packet Switching: Dominant in today’s data centers; involves optical-electrical-optical (OEO) conversion, adding latency and power draw.
- Optical Circuit Switching: Establishes dynamic optical paths between endpoints with no OEO conversion, making it ideal for large-scale, high-volume data transfers typical of AI model training and inference.
Google’s Approach to OCS:
Google has been an early and prominent adopter of OCS within its internal networking architecture. Under its Project Apollo, Google deploys OCS in tandem with its Jupiter/AI networks to optimize interconnect bandwidth among TPU clusters.
Key Industry Developments:
- Coherent Corp has unveiled optical switches digital liquid crystal technology.
- iPronics is commercializing a general-purpose programmable photonic switch based on its SmartLight silicon photonic chip. Unlike fixed-function optical switches, SmartLight allows users to dynamically control switch behavior via software, supporting evolving AI workloads and hybrid topologies.
- Intel, through its silicon photonics group, has introduced OCS reference designs using optical phased arrays (OPA) and low-loss silicon nitride waveguides. Intel aims to bring reconfigurable optical mesh networks to AI training fabrics, where deterministic latency and bandwidth guarantees are essential.
- Lumentum, a co-leader of the OCP OCS Subproject, continues to invest in scalable, low-power OCS platforms based on MEMS.
🌐 Why It Matters: OCS represents a paradigm shift in how networks handle AI workloads. By eliminating electrical bottlenecks and enabling dynamic, high-throughput optical paths, OCS can significantly reduce energy per bit and enable the scale-out needed for trillion-parameter models. As more hyperscalers and vendors commit to open interfaces and collaborative development, OCS may soon become a foundational element of the AI data center stack.
