CIOE 2026 | The 27th China International Optoelectronic Expo || Shenzhen World

Rapid AI iteration is driving the continued expansion of intelligent computing clusters. Traditional electrical switching architectures are struggling with bandwidth bottlenecks, high latency, and surging power consumption. Interconnect has become the primary constraint on AI computing power release. OCS (Optical Circuit Switching) technology, with its revolutionary advantage of switching signals directly in the optical domain—eliminating the "optical-electrical-optical" conversion—breaks through the "bandwidth wall" and "power wall" of electrical switching, positioning itself as a critical enabler for next-generation AI infrastructure.

On July 23, China International Optoelectronic Exposition (CIOE) and C114 jointly hosted the 2026 China Optical Communications High-Quality Development Forum – Optical Switching Systems Technology Session. Industry representatives from telecom operators, internet companies, optical communications vendors, and research institutions engaged in in-depth discussions on the latest OCS technology advances, commercial deployment cases, product iteration roadmaps, and market evolution trends, offering multi-dimensional industry perspectives for the high-quality development of AI computing infrastructure.

Optical Interconnect in AI Data Centers: The Rising Value of OCS

Wang Ruixue, Technical Manager at China Mobile Research Institute's Basic Network Technology Department, noted that AI clusters are experiencing explosive growth, rapidly scaling toward 100,000-card configurations, placing extreme demands on interconnect. Traditional electrical interconnect is approaching physical limits in bandwidth and power consumption, making optical interconnect—with its high bandwidth and low power consumption—the core path forward. In this context, optical switching technology has become a "powerful tool" for enhancing cluster reliability and breaking through scale constraints.

Wang Peng, Optical Network Architect at Alibaba Cloud, pointed out that AI cluster scaling faces four walls: bandwidth, power, cost, and reliability. The industry expects OCS to deliver low-power, low-latency, reconfigurable interconnect capabilities for AI networks, while optimizing system TCO at specific scales. OCS is not intended to replace traditional electrical switching but to evolve in synergy with it, optimizing interconnect efficiency for AI clusters in specific scenarios.

Hu Yakun, Optical Network Researcher at China Unicom, introduced that in Scale‑across scenarios, OCS features—high bandwidth, massive port counts, rate independence, excellent scalability, and topological reconfigurability—align well with DCI network requirements. In Scale‑out scenarios, OCS can establish direct links for high-traffic Pods, allocate network bandwidth by training task, support incremental builds and multi-generation equipment coexistence, and facilitate fault and maintenance scheduling. In Scale‑up scenarios, OCS enables the formation of supernodes of varying sizes per task, enhancing compute power through optical interconnect.

Liu Yuyang, Senior Engineer at China Telecom Research Institute, stated that AI inference is shifting from offline invocation to real-time online services. Low latency, high concurrency, and proximity to end-user experience are driving compute power to extend from central AIDCs to metro and edge locations. Carriers' CO (Central Office) nodes, located close to users and equipped with facility space, power, fiber, OTN, ROADM, and O&M systems, are natural hosts for edge AI inference. OCS enables fiber-level/optical-path-level reconfiguration, merging multiple metro rings on demand into larger logical topologies, achieving pooling and elastic scaling of multi-CO compute resources, thus forming larger metro-scale AI inference compute pools.

Domestic Supply Chain Accelerating Maturity

Regarding the OCS domestic supply chain, Alibaba Cloud's Wang Peng noted that China has full control over optical components, LC connectors, and system software. Partial manufacturing capabilities exist for MEMS micromirrors, silicon photonics chips, DLC panels, precision packaging, driver ASICs, FPGAs, and test/calibration equipment, though yield and consistency still require improvement. DBS technology, involving deep coupling of piezoelectric actuators, precision packaging, and closed-loop servo circuits, lacks domestic core capabilities. Therefore, the current viable path is to focus on the MEMS and silicon photonics routes, where domestic technology is relatively strong.

Lyu Mingyang, Senior Principal Analyst for Optical Components at Omdia, advised new entrants to carefully assess their positioning: rather than pursuing complete system solutions, it may be wiser to enter the supply chain as core component suppliers, or to form deep partnerships with target customers (such as domestic computing power providers) to jointly plan for application scenarios targeting larger-scale node interconnection beyond 2028.

Su Guoyan, Sales Director at Guilin GLsun Science and Tech Group Co.,LTD, introduced the company's OCS system solution, built on the mainstream MEMS micromirror array technology path. The solution employs dual-axis MEMS micromirror arrays as the core optical switching device, enabling millisecond-level dynamic optical path scheduling and non-blocking switching, with high reliability and rapid reconfiguration capabilities—providing the core hardware foundation for elastic high-speed optical networks. GLSUN has established multiple phased goals and will continue driving OCS product innovation and solution implementation, positioning itself as a core engine for AI computing optical interconnect.

Zhang Hua, CTO of the Fiber Optic Devices and Instruments Division at LUSTER LightTech, introduced that the DBS (Direct Beam Steering) solution based on piezoelectric ceramics achieves precise rotation of fiber collimators via 2D piezoelectric actuators, enabling optical signal switching through direct spatial coupling alignment. Compared to the MEMS micromirror array approach, DBS technology has no mechanical wear, offering significant reliability advantages. LUSTER, in partnership with H+S Polatis, has deployed OCS products with over 18.8 billion port-hours of cumulative operation.

Liang Jialun, Product Manager at TRIPLE-STONE, noted that the value of OCS has been validated through exploration by AI cloud providers and the industrial chain. With AI driving continuous expansion of computing clusters and clear near-term demand, the industry discussion on OCS has shifted from "whether it's valuable" to "whether it can be engineered for deployment." TRIPLE-STONE has already developed product portfolios covering 64×64, 128×128, 256×256, and 300×300 configurations.

Liang emphasized that engineering is not a single-point breakthrough but a complete chain from definition, architecture, and prototyping to iteration and delivery. Tri-Sun has distilled four key lessons: first, shifting focus from device performance to system consistency; second, calibration must be not only accurate but also fast and fully automated; third, long-term stability means the system can handle change, not just maintain constant conditions; and fourth, mass production is not about replicating prototypes but about controlling distribution.

Driving Commercial Deployment Through Collaborative Efforts

Alibaba Cloud's Wang Peng stated that OCS commercial-scale deployment still relies on coordinated efforts among four parties: cloud service providers, equipment/device vendors, module vendors, and standards organizations.

Cloud service providers need to drive demand and validate scenarios, defining deployable requirements covering operability, monitoring, and batch deployment. Equipment and device vendors must push the engineering maturity of MEMS/SiPh/DLC and other technology paths, achieving high port count, low insertion loss, low cost, and high reliability. Optical module vendors need to adapt to 800G/1.6T optical interconnect technologies and link budgets, promoting compatibility across pluggable, NPO, and CPO solutions. Standards and open-source organizations must drive standardization of interfaces, alarms, and management models, while refining test methods, interoperability specifications, and certification systems to lower cross-vendor integration barriers.

China Mobile's Wang Ruixue noted that introducing OCS to AI data centers can yield cost and power consumption benefits. However, its slower switching speed and limited reconfigurability entail complex topology engineering and traffic engineering investments. Additionally, traffic rerouting can lead to reduced effective compute power and increased out-of-order processing complexity. Therefore, it is essential to advance R&D on new optical switching equipment, intelligent topology reconfiguration, and opto-electronic protocol stack coordination, while simultaneously promoting standards development and industrial collaboration across all layers to enable the deployment of integrated opto-electronic networking solutions.

Wang Ruixue called for industry collaboration to advance standards and specifications, building a solid foundation for OCS deployment in AI computing scenarios: first, establishing a unified standards system to anchor technology evolution; second, standardizing core interface protocols to break down heterogeneous interoperability barriers; and third, deepening industrial ecosystem collaboration to accelerate deployment validation loops.

As Alibaba Cloud's Wang Peng observed, OCS currently lacks large-scale deployment precedents in China, and the journey from lab to production environment remains substantial. In terms of commercialization readiness, device breakthroughs are just the starting point—system engineering, O&M systems, and ecosystem collaboration collectively determine the pace of commercialization. Domestic end-user demand, system software, and system engineering still require time for practical refinement, and uncertainties remain. End users must serve as demand-side drivers, uniting industry stakeholders to propel OCS from "viable" to "large-scale deployment," building a solid technical foundation for AI cluster interconnect.

From Proof-of-Concept to Mass Adoption: How Industry Collaboration Is Scaling OCS