Looking ahead to 2026, the "siphon effect" of AI is expected to be significant, potentially driving continued global AI resonance. Overseas, tech giants like Google are consistently raising their AI capital expenditure guidance, and the token consumption of large models such as Google's Gemini has surged significantly, gradually highlighting the positive feedback loop of AI. Domestically, major Chinese AI companies may enter a phase of large-scale investment in AI computing power. The firm is optimistic about five core themes: "optics, optical fiber, liquid cooling, domestic AI, and satellites," while also recommending a focus on sectors such as operators and AI applications. The main views of Kaiyuan Securities are as follows:
Theme for 2026: Global AI Resonance, Focusing on Four Main Areas: "Optics, Optical Fiber, Liquid Cooling, and Domestic AI" 2026 may mark the inaugural year for the mass adoption of 1.6T optical modules, with silicon photonics technology likely accelerating its implementation and potentially becoming the mainstream solution for high-speed 1.6T optical modules. This development could significantly boost demand for optical chips such as EML and CW lasers. The triple resonance of "AIDC + drones + DCI" is expected to continue driving demand for optical fibers and cables. Concurrently, multiple new technologies, including OCS/CPO/hollow-core fiber/thin-film lithium niobate/LPO/LRO, are evolving in parallel, potentially fostering new investment opportunities. In 2026, with the gradual shipment of Nvidia's GB300 NVL72, Rubin, and Rubin CPX series cabinets, increased shipments of the latest computing chips from ASIC giants like Google, the scaled shipment of domestic computing chips, and the gradual ramp-up of domestic super nodes, the penetration rate of liquid cooling is likely to rise substantially. Driven by national policies and domestic tech giants, the domestic AI industry chain is poised for robust growth in 2026, with a recommended focus on three core directions: computing, networking, and AIDC.
Theme for 2026: Domestic AI and Satellites Jointly Advance, Building an Integrated Space-Air-Ground Computing Network The demand for large model training continues to grow, and space, with its unique advantages such as unlimited solar energy, natural vacuum cooling environment, and seamless global coverage, has become a core direction for the next generation of computing power development. SpaceX plans to deploy a massive space computing constellation with millions of satellites, while companies like Blue Origin, Google, and Starcloud are also making strategic moves. China places high importance on the construction of the space-air-ground computing network, with satellites serving as the foundational cornerstone of the space network. It is recommended to focus on four key segments: (1) Satellite manufacturing: Benefiting primarily from the incremental demand for satellite launches, pay attention to satellite payload suppliers and satellite platform component suppliers. (2) Satellite launch: Focus on industry catalysts from launch resource allocation, launch pace, and technological advancements. (3) Ground equipment: Pay attention to various aspects of high-value core network construction and the terminal market. (4) Satellite operations: With Star Network and Yuanxin leading the Star Network and G60 projects respectively, progressing on dual tracks is expected to rapidly build the satellite network.
Theme for 2026: AI Applications Surge, Creating Opportunities for Operators, Modules, Controllers, and CDN In the AI era, operators are vigorously developing AI, and AI cloud services are expected to become a new engine for performance growth. Advances in AI model-based technologies like speech recognition and image recognition are accelerating the development of AI toys. These toys achieve anthropomorphic interaction capabilities by integrating hardware such as intelligent computing modules and cellular/Wi-Fi modules. As edge AI models continue to optimize, demand for intelligent computing and communication modules is likely to accelerate. AI models form the foundation for embodied intelligence development, with smaller models after pruning and quantization performing impressively on edge terminals, potentially gradually increasing demand for intelligent computing modules and controllers. AI applications are expected to generate demand for millisecond-level responses. CDN, through its edge node network deployment, can bring computing power closer to end-users, enabling localized and lightweight data processing, and AI applications may continuously drive demand for CDN.
Risk warnings: AI development falling short of expectations, slower-than-expected satellite development, and escalating Sino-US trade frictions.
Comments