AMD Nvidia Broadcom Create Optical Interconnect for AI Clus

AMD, Broadcom, and Nvidia have collaborated with hyperscalers Microsoft, Meta, and OpenAI to launch the Optical Compute Interconnect (OCI) Multi-Source Agreement group. This initiative aims to define a unified optical connectivity standard for scale-up interconnections within large AI clusters, transitioning from traditional copper cables to optical solutions for improved performance. The technology roadmap targets speeds scaling up to 3.2 Tb/s per fiber, utilizing techniques such as NRZ signaling and wavelength-division multiplexing (WDM). The effort focuses on developing a common optical physical layer (PHY), thereby allowing interoperability between various processing units and protocols, including UALink and NVLink.

This collaboration is significant as it addresses a critical bottleneck in AI infrastructure, ensuring that hyperscalers can connect multiple accelerators effectively and sustainably. By standardizing optical interconnects, the OCI aims to reduce system integration complexities and accelerate deployment cycles for next-generation AI hardware. Furthermore, as the ecosystem evolves, this initiative is expected to support numerous pluggable optical modules and enhance AI data processing capabilities, ultimately increasing national AI autonomy and reducing reliance on foreign technology providers.