A 19-company coalition has officially launched the Open Silicon Photonics for AI Systems initiative as a formal workstream within the Open Compute Project (OCP). Led by Lightmatter, the group published a 300-page foundational white paper to standardize silicon photonics-ready infrastructure. This move aims to address the physical limitations of copper interconnects, providing a shared architectural blueprint to scale AI clusters from 72 to over 1,024 nodes while ensuring multi-vendor interoperability.
19-Company Coalition Standardizes AI Infrastructure
The initiative includes 10 founding members and nine new additions, such as Dell Technologies, Qualcomm, Celestica, Flex, and Foxconn Interconnect Technology. By establishing the "Architecture Vision: Open Silicon Photonics for AI Systems" white paper, the coalition provides a framework for scaling Modular Hardware System (MHS)- and Open Rack v3 (ORv3)-compliant clusters. This collaborative effort seeks to move co-packaged optics (CPO) from early deployments toward mainstream AI infrastructure. The group aims to support massive 100,000-XPU clusters by creating a technology-agnostic approach. This architecture supports diverse photonic solutions, including silicon photonics, VCSELs, and micro-LEDs, which helps foster a robust, multi-vendor supply chain capable of meeting extreme performance and power demands.
Overcoming Copper Constraints via Photonic Architectures
As SerDes rates approach 448G, the physical reach of copper is shrinking to tens of centimeters, creating a bottleneck for large-scale AI systems. The OCP reference architecture addresses this by facilitating a transition to all-optical fabrics. This shift is necessary because interconnect performance has become as fundamental to AI infrastructure as raw compute power. The initiative provides hyperscalers with a scalable framework that remains interoperable across various SerDes generations and XPU design cycles. By optimizing energy efficiency and data transfer speeds, the framework helps mitigate the constraints that copper imposes on system scalability and accelerator utilization. The formal process for this initiative includes the anticipated submission of the first technical specifications in Q4 2026.
Key Takeaways
- The initiative includes 19 companies, including Dell Technologies, Qualcomm, and Lightmatter, operating within the Open Compute Project.
- The foundational white paper provides a blueprint to scale AI clusters from 72 nodes to over 1,024 nodes.
- Technical specifications for the new architecture are expected to be submitted in Q4 2026.
TechInsyte's Take
In our view, this initiative signals a critical industry pivot from compute-centric to interconnect-centric scaling strategies. As copper reaches its physical limits at 448G SerDes rates, the move toward silicon photonics is no longer optional for hyperscalers. By standardizing through the OCP, these 19 companies are attempting to prevent vendor lock-in during a period of massive infrastructure transition. This collective approach to CPO-enabled architecture is a strategic move to ensure that the next generation of 100,000-XPU clusters remains economically and physically viable.
Questions & Answers
How does this initiative address the physical limitations of current AI hardware?
The initiative addresses the shrinking reach of copper interconnects, which is currently limited to tens of centimeters as SerDes rates approach 448G, by providing a framework for transitioning to all-optical fabrics.
What is the projected timeline for technical standardization?
The coalition expects to submit the first formal specifications for the Open Silicon Photonics for AI Systems initiative in Q4 2026.
Which hardware standards will the new CPO blueprint support?
The shared CPO blueprint is designed to scale Modular Hardware System (MHS)- and Open Rack v3 (ORv3)-compliant AI clusters.
Why is multi-vendor interoperability a focus for this coalition?
The coalition uses a technology-agnostic approach to support diverse photonic solutions like silicon photonics and micro-LEDs, aiming to foster a robust, multi-vendor supply chain for massive XPU clusters.
Source: Businesswire
