Co-packaged optics (CPO) has gained attention recently due to its power efficiency in data centers. While most leading proponents of CPO targeting networking applications have discontinued CPO programs due to macroeconomic headwinds, the situation of CPO for artifical Intelligence (AI) and machine learning (ML) systems is different. AI models have an insatiable demand for computing power, storage, and data movement, and traditional architectures are becoming the main bottleneck for scaling ML. As a result, new optical interconnects have emerged for HPC and new disaggregated system architectures, which Yole Intelligence, part of Yole Group reviews in its new report, Co-packaged Optics for Datacenter. In-package optical I/O technology for xPUs, memory, and storage can help achieve the necessary bandwidths. Moreover, the potential for billions of optical interconnects in the future is driving big foundries to prepare for mass production (including silicon photonics process flows) of any PIC architecture from design houses. CPO revenue was about US$38 million in 2022 and is expected to reach US$2.6 billion in 2033, at a 46% CAGR for 2022 – 2033, driven by accelerated data movement in AI/ML gear.
Comments