Workflow
中信证券:OCS技术路线未来增长确定性较强 关注三大投资主线
智通财经网·2025-08-27 01:25

Core Insights - The report from CITIC Securities indicates that the use of OCS in AI data centers is still in its early stages, with Google having already implemented products while most manufacturers are actively developing their technologies [1] - CITIC Securities believes that the OCS technology route is highly feasible with significant advantages, leading to strong growth certainty in the future [1] - The report outlines three investment themes across the entire industry chain: focusing on manufacturers with core optical communication equipment in the complete machine segment, optical MEMS suppliers and foundries in the core component segment, and domestic optical component suppliers in the optical component segment [1] Investment Opportunities - In the complete machine segment, the market space is expected to exceed $40 billion in the long term, with both specialized OCS manufacturers and traditional optical communication equipment manufacturers being able to undertake AI OCS-related business [3] - In the core component segment, the market space for WSS based on MEMS or LCoS technology is projected to reach several billion dollars, with MEMS currently benefiting mainly foundries due to product immaturity, while LCoS manufacturers may expand into the optical communication field [3] - In the passive component segment, OCS increases the usage of optical isolators, micro-lenses, and beam splitters, with the market space for optical isolators potentially reaching $5 billion and the global competitive landscape being relatively fragmented [3] Technological Developments - The AI era necessitates new networks, with OCS addressing the bottlenecks of Scale out, providing high bandwidth, low latency, and low-cost network solutions compared to previous technologies [2] - OCS technology is expected to penetrate AI clusters further, replacing spine switches, leaf switches, and even taking on inter-card connections and on-chip networks, thereby expanding market opportunities [2] - The demand for AI large models requires a new network architecture characterized by large-scale global data exchange, which is fundamentally different from previous data center network architectures [1]