The use of LPO and CPO in scaling networks will bring double-digit growth to this market in 2028-2030.
Nvidia's business growth was the primary factor impacting optical transceiver sales in 2023-2024. Nvidia's new AI cluster designs require a large number of transceivers. Until 2023, Nvidia's AI cluster systems used only copper cables and active optical cables (AOCs). InfiniBand NDR (400G) connections utilize 400/800G SR4/SR8 and DR4/DR8 pluggable transceivers instead of AOCs. The latest XDR systems will use 1.6T DR8 and 2xFR4 pluggable connectors, including some LPOs. Nvidia is also actively developing CPOs, which will be the best option for extending NVLink connectivity over fiber. We expect both LPOs and CPOs to be deployed in scaling networks starting in 2026-2027, reaching high volumes in 2028.
This report explores the evolving role of optics in AI clusters, encompassing both connectivity and switching. It presents sales data for optical transceivers for compute nodes and AI clusters in cloud data centers for 2021–2024, with a forecast for 2025–2030. Importantly, the forecast includes detailed models for the deployment of optics in AI clusters (back-end networks) at Alibaba, Amazon, Google, Meta, Microsoft, and Oracle, illustrating the differences between the optics used in compute nodes (front-end networks). A detailed analysis of the critical role of optical connectivity, including NVLink, UALink, and CXL/PCIe, in the implementation of AI cluster architectures is central to this report.
