Guolong Technology Deepens OCS Strategy

C114 News, July 24 (Nanshan) — Driven by AI, computing infrastructure continues to expand, with clusters growing from thousands to tens of thousands and now hundreds of thousands of GPUs. OCS optical switching has become a hot topic of industry research. At the online seminar "Optical Circuit Switching (OCS): Rebuilding the Ultra-High-Speed Interconnect Foundation for AI Clusters," co-hosted by CIOE and C114, Su Guoyan, sales director of Guanglong Technology, delivered a presentation sharing the underlying logic behind OCS's rapid growth, product solutions and technological innovation, and discussed future development trends.

Su Guoyan first noted that three core bottlenecks constrain the release of AI computing power. First is the "Power Wall": traditional electrical switching architectures involve frequent optical-electrical-optical conversions, causing enormous energy losses, with network power consumption accounting for 20-25% of total data center power usage, making it a key bottleneck for energy-efficiency optimization. Second is the "Latency Wall": each hop through a traditional electrical switch introduces hundreds of nanoseconds of processing delay, and the accumulated latency in multi-hop networks significantly reduces synchronization efficiency in distributed AI training, limiting computing performance. Third is the "Cost Wall": ASIC chip iteration cycles are short, and network speed upgrades often require replacing equipment across the entire network, bringing not only high capital expenditure (CAPEX) but also greatly increased operational complexity and cost.

As AI computing expands, optical communications are rapidly becoming foundational AI infrastructure. 800G optical transceivers have become standard equipment in AI computing centers, 1.6T products are already in volume production, and the technology iteration cycle has been compressed from 4-5 years to 2 years, ushering in explosive capacity expansion and rapid technological turnover across the industry. Meanwhile, led by AI cloud providers, the growth logic has shifted from simple penetration gains to a dual drive of per-port value and computing demand, achieving a deep leap from "scale expansion" to "value upgrading" and reshaping the industry's competitive landscape.

In this process, OCS is evolving from an exclusive solution used by a few AI cloud providers such as Google into general-purpose infrastructure for AI computing. Su Guoyan explained that OCS currently has three mainstream technology routes. First is the MEMS micromirror array, a relatively mature technology featuring high channel count, low loss and millisecond-level switching, and one of today's mainstream market solutions. Second is liquid-crystal electro-optic control, characterized by high reliability and stability, no mechanical wear and extremely long service life, making it the preferred choice for applications demanding high reliability. Third is silicon photonics integration, offering ultra-fast response and high integration, and pointing toward the future of high-integration, low-latency solutions.

With deep expertise in core optical interconnect technologies, Guanglong Technology has developed an OCS system solution based on the mainstream MEMS micromirror array technology route. The solution uses a two-axis MEMS micromirror array as the core optical switching device, enabling millisecond-level dynamic optical path scheduling and non-blocking switching. With high reliability and fast reconfiguration capability, it serves as the core hardware foundation for building elastic, high-speed optical networks.

Su Guoyan said the solution offers core advantages including extremely low insertion loss, ultra-high isolation, millisecond-level switching and high-density integration: minimum insertion loss of 1 dB, channel isolation exceeding 50 dB, switching response under 30 milliseconds, and support for 128✖️128 ports (4U rack). These advantages make the solution applicable to data center interconnect (DCI) and backbone network optical switching scenarios, enabling low-latency, high-bandwidth, highly flexible elastic optical internet networks. Meanwhile, Guanglong Technology's robust supply chain, flexible manufacturing and volume production capabilities provide customers with stable, reliable assurance for deployment.

Su Guoyan concluded that as OCS technology and the supply chain mature, commercialization will accelerate across the board and application scenarios will continue to expand. O-Net Technologies has set phased targets and will keep advancing OCS product innovation and solution deployment, aiming to build a core engine for optical interconnect in AI computing.