Given the surge in large models sparked by ChatGPT, AI computing power demand is growing exponentially. Traditional data center networks are facing bandwidth bottlenecks, energy consumption crises, and scalability challenges, quietly triggering a fundamental infrastructure revolution from "electrical switching" to "optical switching." As a next-generation interconnect technology that performs optical path switching directly without optical-to-electrical conversion, OCS is becoming the focus of tech giants such as Google, Microsoft, and NVIDIA.
01. OCS Resolves the Dilemma of Traditional Switching in Data Centers
OCS (Optical Circuit Switch) is a technology that directly switches optical signals between fiber ports without optical-to-electrical/optical-to-optical (O/E/O) conversion. It is applied in scenarios such as AI computing clusters, leaf-spine architecture interconnections in hyperscale data centers, and high-speed communication in supernode clusters. By completing the routing and switching of data signals directly in the optical domain without O/E/O conversion, OCS fundamentally avoids the bandwidth bottlenecks and power consumption issues of traditional electrical switching in high-speed transmission, significantly reducing signal transmission latency. Industry technology research and practical implementation have verified that OCS technology can help reduce the overall power consumption of AI computing clusters and data center optical interconnect systems by more than 30%.
Currently, data centers using traditional OEO (Optical-Electrical-Optical) switches face several critical challenges in practical applications.
Comparison Between OEO and OCS
Source: Compiled by MIR Research based on public information
In recent years, the OCS industry has developed rapidly both domestically and internationally. Overseas, the Open Compute Project Foundation established the OCS sub-project in July 2025, co-led by companies such as Lumentum and iPronics to promote industry standardization collaboration. *The Open Compute Project (OCP) is an open-source hardware collaboration initiative launched in 2011 by Facebook, along with Intel, Rackspace, and others, aiming to improve industry efficiency by sharing data center and server designs.
In March 2026, NVIDIA announced multi-year agreements with optical technology companies Lumentum and Coherent, investing $2 billion in each company. The non-exclusive cooperation with Coherent includes billions of dollars in procurement commitments and access to advanced laser component production capacity, while also supporting Lumentum's R&D, production capacity, and operations to help it build a new wafer fab in the United States.
Domestic progress is equally active. In October 2025, China Mobile Cloud released a report considering the use of OCS to replace the original Super Spine layer in the future, promoting the transformation of the computing network architecture towards "all-optical native." In the same month, the Ministry of Industry and Information Technology (MIIT) deployed the "Millisecond Computing" special action, explicitly promoting the deployment of OCS and other technologies in computing centers to optimize network latency. The dual drive of policy and industry paves the way for the large-scale application of OCS.
02. Four Mainstream OCS Technology Routes Running in Parallel
Currently, the OCS industry has not yet formed unified technical standards. Four major routes—MEMS, liquid crystal, piezoelectric ceramics, and silicon photonics waveguides—coexist, competing in terms of cost, performance, and reliability.
Currently, Google-led MEMS technology entered the commercial stage in 2025, capturing over 90% of the market share with an annual production of nearly 10,000 units, while other technologies remain in the market validation stage with small batches of dozens of units. It is expected that Google's MEMS solution shipments will exceed 10,000 units in 2026, adopting two supply models: first, independent design with designated component suppliers and contract manufacturers completing the whole machine assembly; second, direct procurement of whole machine solutions, with cooperating suppliers including Lumentum and Coherent.
Comparison of Advantages and Disadvantages of the Four Technology Routes
Source: Cignal AI
03. OCS Targets Three Major AI Data Center Scenarios
OCS is not just a concept confined to the laboratory; its applications are rapidly materializing alongside the expansion of AI computing clusters, primarily focusing on three core scenarios:
The first major scenario is intra-rack interconnection for AI chip clusters, exemplified by Google's TPU clusters. *TPU (Tensor Processing Unit) is an Application-Specific Integrated Circuit (ASIC) custom-developed by Google to accelerate machine learning workloads. Today, TPUs support most of Google's AI services.
Starting with TPUv4, Google introduced OCS to build large-scale Superpods. In a cluster composed of 4,096 TPUv4 chips, Google used 48 OCS switches. For the latest Ironwood, its super cluster of 9,216 TPUs doubles the port demand for OCS, requiring 48 OCS units with 600 ports each. Here, OCS achieves stable, high-bandwidth, and low-latency physical connections between chips, serving as the foundational network cornerstone for supporting 10,000-card-level intelligent computing clusters.
Google 4096 Supernode Architecture
Image Source: Google
The second major scenario is replacing data center Spine layer switches. In traditional Clos network architectures, Spine layer traffic is relatively stable and predictable, which is exactly where OCS excels. Google introduced its self-developed Apollo OCS in its Jupiter data center network, successfully reducing flow completion time by 10%, increasing throughput by 30%, reducing power consumption by 40%, and cutting costs by 30%. This proves that OCS has significant performance and energy efficiency advantages in data center backbone networks.
The third major scenario is Data Center Interconnect (DCI). As individual data centers approach their limits in power and capacity, connecting multiple distributed data centers into a "computing power factory" becomes inevitable, which is the Scale-Across concept proposed by NVIDIA. The DCI scenario has extremely high requirements for bandwidth, distance, and energy consumption, and the high bandwidth, rate-independence, and low power consumption characteristics of OCS highly match these requirements. Vendors such as Coherent have explicitly announced plans to launch C-band OCS products for DCI in 2026.
Coherent Expected to Launch OCS for DCI in 2026
Image Source: Coherent
04. Chinese Manufacturers Deeply Integrate into the Core Supply Chain
The booming prospects of the OCS market have attracted a series of manufacturers, ranging from optical component suppliers and whole machine solution providers to traditional optical communication leaders, and the supply chain landscape is rapidly taking shape.
Overview of Chinese OCS Industry Chain Companies
Source: Official websites of respective companies and OFC official website
Dekeli
Wuxi Dekeli Optoelectronic Technology Co., Ltd., formerly known as Wuxi ZTE Optoelectronic Technology Co., Ltd., is primarily engaged in the R&D, production, and sales of optical transceiver modules, optical amplifiers, and optical transmission subsystems. It has formed a complete product layout of "components + modules + systems," while also deploying cutting-edge technologies and products such as high-speed optoelectronic transceiver chips, OCS optical line switching, and DCI data center interconnection. Its technical advantages are mainly reflected in the fields of high-speed and long-distance optical transmission.
In 2025, Dekeli achieved a total operating revenue of 934 million RMB, a year-on-year increase of 10.99%; net profit was 73.1686 million RMB, a year-on-year decrease of 27.16%. During the reporting period, affected by the slowdown in structural demand in the telecom transmission market and intensified competition for some products, the company adjusted its pricing strategy for mature product lines, resulting in a year-on-year decline in product gross profit margin and temporary pressure on operating performance.
In 2025, Dekeli achieved remarkable R&D results, with a breakthrough in OCS and hollow-core fiber amplifiers. The company's silicon-based OCS received overseas sample orders, and the R&D of the second-generation high-dimensional OCS is accelerating (targeting a prototype in H1 2026). Optical amplifier technology continues to achieve breakthroughs, with broadband products, especially L++ products, maintaining market leadership. It has achieved mass shipment of SOA single-channel amplifiers and launched a C-band multi-wavelength amplifier dedicated to hollow-core fiber (output power > 2W) to seize the emerging scenario market.
Accelink Technologies
Wuhan Accelink Technologies Co., Ltd. (Accelink) possesses vertical integration capabilities from chips, components, and modules to subsystems. Through independent R&D and acquisitions, it has formed a layout of three major optical chip platforms (Planar Lightwave Circuit, III-V compound, and SiP (System in Package)). It has a high self-supply rate for low-speed optical chips and is actively developing high-speed products above 100G, achieving vertical industrial integration from chips to subsystems. At OFC 2024, Accelink innovatively launched the latest MEMS series product, OCS (Optical Circuit Switch), which can significantly reduce data transmission latency and data center energy consumption, and greatly shorten the upgrade and construction cycle of data centers.
At the commercialization level, the OCS business has become Accelink's second growth curve. The 2025 interim report shows that the OCS optical switch business is accelerating its implementation, with OCS-related revenue exceeding 180 million RMB in the first half of the year, a year-on-year increase of over 300%. According to public information, Accelink's 192×192 port MEMS-OCS product has passed Google's certification and entered Huawei's supply chain, with a gross profit margin as high as 52%. Accelink has also jointly developed silicon-based MEMS and silicon nitride waveguide heterogeneous integration technology with the Institute of Semiconductors, Chinese Academy of Sciences, aiming to reduce the overall power consumption of OCS from the industry average of 8W/port to 5W/port, with mass production expected in 2026.
Conclusion
Currently, the OCS market is accelerating its transition from the technology validation phase to the large-scale commercialization phase. The MEMS route, leveraging Google's ecosystem dominance, captures over 90% of the market share, making the technological landscape difficult to shake in the short term. At the industry chain level, overseas giants have locked in their first-mover advantages by binding with tech giants, while Chinese manufacturers have deeply integrated into the supply chain in core components and whole machine segments. Companies such as Dekeli and Accelink have achieved breakthroughs in overseas orders or certifications, proving that domestic solutions possess international competitiveness. As AI computing clusters expand towards the 10,000-card level and DCI demands are released, OCS will penetrate more data center scenarios from the "exclusive configuration" of supernodes, providing critical support for cost reduction and efficiency improvement in computing infrastructure.
This article is written based on the report information from the MIR DATABANK database.
*Log in to the MIR DATABANK Professional Manufacturing Data Platform to get more manufacturing data information, company information, and research reports! https://www.mirdatabank.com