A 19-company industry coalition has formally launched the Open Silicon Photonics for AI Systems initiative within the Open Compute Project (OCP), aiming to establish an open architecture for deploying co-packaged optics (CPO) across large-scale AI infrastructure.
Led by Lightmatter, the initiative brings together companies including Celestica, Dell Technologies, Flex, Foxconn Interconnect Technology, Global Unichip Corporation, Hyve Solutions, Keysight, Qualcomm and Quanta Cloud Technology.
The group has published a foundational architecture white paper designed to guide development of a shared CPO blueprint for Modular Hardware System (MHS) and Open Rack v3 (ORv3) platforms.
The architecture is intended to support AI clusters scaling from 72 to more than 1,024 nodes while enabling multi-vendor interoperability.
The initiative comes as increasing AI accelerator density and rising SerDes speeds place greater pressure on conventional electrical interconnects.
With SerDes rates approaching 448G, the coalition says copper reach is becoming increasingly constrained, accelerating the transition towards optical connectivity within AI systems.
A key objective is to create a technology-agnostic infrastructure that can accommodate different photonic technologies and successive generations of optical engines without requiring major changes to the surrounding rack infrastructure.
The proposed architecture is also intended to support both scale-up and scale-out networking, including protocols such as NVLink and UALink, while allowing multiple fabric planes to coexist on common passive optical infrastructure.
“Interconnect has become as fundamental to AI infrastructure performance as compute,” said Nick Harris, founder and CEO of Lightmatter. He added that the white paper provides the foundation for developing a shared MHS-based CPO rack architecture compatible with hyperscaler deployment models.
The coalition expects the first specifications emerging from the initiative to be submitted in the fourth quarter of 2026, as the industry works towards moving CPO from early deployments into more widely adopted AI infrastructure.
