Guoxin Securities: The AI industry chain is moving from differences to consensus in the performance trend 2026 is expected to be a year for local hard technology to be harvested

The Zhitong Finance App learned that Guoxin Securities released a research report saying that looking ahead to 2026, the reasoning capabilities of big AI models will continue to be iterated, a closed loop between large models and end-side applications is being formed, and the trend of oversupply at the computational power+memory hardware level will continue, while the expansion progress of advanced domestic processes and the pace of autonomous and controlled progress are still far from expected. 2026 is expected to be “the year of the harvest of local hard technology from the flames of the stars to the full spread of the earth”.

Guoxin Securities's main views are as follows:

In 2025, the AI industry chain moved from differences to consensus in performance trends. 2026 is expected to be a year for local hard technology to be harvested

The bank previously expressed its opinion that the electronics industry's boom cycle has declined for nearly 2 years since 2021, bottoming out in the second half of 2023. With the return of the Huawei Mate series as a landmark event, it is still in a moderate upward process driven by AI innovation. The bank previously expressed its opinion that in the process of “macroeconomic policy cycle, industrial inventory cycle, and AI innovation cycle” resonating upward, the industry showed a significant trend of valuation expansion with the help of the rapid expansion of passive funds. After experiencing the narrative logic shock of “computing power deflation” caused by the rise of Deepseek in 2Q25 and the impact of the tariff war initiated by the US, the market has gone from disagreement to consensus in the outstanding performance trend of the AI industry chain since 3Q25. As of December 16, electronics had risen 40.22%, ranking third in the industry. The bank believes that the Chinese technology industry, which is already trending in the 5G innovation cycle that fully began in 2020, is showing stronger global competitiveness in a new round of AI innovation cycle after facing America's “sanctions policy” and experiencing rapid accumulation of “human goods” for more than 5 years. 2026 is expected to be “the year of harvesting local hard technology from the blaze of stars to the full spread of local hard technology.”

AI models are competing, Nvidia is leading the iteration of computing power, and the PCB and server industry chains continue to grow at a high rate

Thanks to the architectural innovation of big models: using hybrid expert architectures to achieve more efficient reasoning through sparseness, using innovative attention mechanisms to reduce computational complexity and memory requirements, and multiple rounds of deduction in deep thinking mode to reduce illusions, etc., big models at home and abroad have achieved continuous progress in multi-modal understanding, reasoning, and AI applications. Benefiting from the expansion of demand for computing power such as CSP and sovereign cloud, and the booming development of AI inference applications, TrendForce expects the total capital expenditure of the world's top eight CSP to increase by 40% to 600 billion US dollars+ in 2026, and global AI server shipments will increase 20.9%. At the same time, following the upgrade from GB200 to GB300 in 2025, Nvidia's next-generation Rubin-architecture AI server will revolutionize discrete reasoning in 2026. Nvidia expects that by the end of 26, the total number of Blackwell and Rubin GPUs will be shipped to reach 20 million units, and the total order will reach 500 billion US dollars. Based on the expansion of the market size of the computing power arms race and the ASP increase brought about by the iteration of computing power products, along with the expansion of intelligent computing clusters brought by Scale Up and Scale Out, the bank believes that in 2026, the PCB and server industry chain, which is deeply involved in the division of labor in the global industrial chain, including related upstream materials and liquid cooling and cooling, will usher in a period of rapid growth in volume and price. Focus on: IFF, Shenghong Technology, Shengyi Technology, Huaqin Technology, Shanghai Electric Power Co., Ltd., Lansi Technology, Lixun Precision, Pengding Holdings, Dongshan Precision, and Jingwang Electronics.

Computing power+memory: Domestic general-purpose computing power chips and ASIC solutions work together, and the price increase is expected to continue throughout the year when the memory is out of stock

In terms of computing power, domestic chips are currently being actively updated and iterated. Huawei plans to launch the Shengteng 950PR in 1Q26 and the Atlas 950 SuperPod supernode in 4Q26; domestic cards such as Cambrian, Mu Xi, Wall, and Moore Thread have been successfully introduced into the Intelligent Computing Center. At the same time, limited by the US BIS's repeated sanctions on domestic chips, CSP companies' compliant ASIC projects will simultaneously usher in development opportunities. Among them, self-developed projects from non-first-tier cloud manufacturers are expected to bring significant increases to domestic ASIC manufacturers, and storage solutions will also become a key point of differentiated ASIC competition. In terms of storage capacity, DRAM in the AI era gradually changed from a “subsidiary role” to a “performance bottleneck breakthrough”, and will grow at an accelerated pace driven by supply-side structural adjustments and AI demand. Demand for DRAM bits is expected to increase 26% year-on-year in 2026. At the same time, with the rise of AI reasoning, the limitations of traditional HDDs in terms of read/write speed, response delay, and energy efficiency have accelerated SSD penetration, and the NAND shortage trend has spread from local applications to the whole market. The price index has risen by more than 40% from September to December '25. The bank believes that DRAM and 26 NAND will continue to show severe supply shortages, and prices are expected to continue to rise. Focus on the industrial chain related to computing power and memory: Cambrian, Aojie Technology, VeriSilicon, Demingli, Jiang Bolong, Zhaoyi Innovation, Beijing Junzheng, Weidu Technology.

Capacity+electricity: Capacity has become an important breakthrough in improving computing power, and the increase in computing power drives the simultaneous upgrading of power architectures

The capacity process must not only solve the problem of data entering and leaving memory, but also achieve smooth communication within servers, between racks, and clusters. In the context of limited domestic high-end computing power chip flow, the optimization of the capacity chain has become an important breakthrough. It is estimated that in 2024-2030, the global high-speed interconnect chip market CAGR will be 21.2%, and the share of the Chinese market will increase from 25% to 30%, creating a broad incremental market for MRDIMM memory, PCIe interconnect chips, CXL interconnect chips, silicon chips, OCS and other industrial chains, focusing on LanGuang Technology and Blue Light Technology . At the same time, as the power consumption of data center chips increases and the processing power of racks soars, Infineon predicts that the power consumption of a single GPU will increase exponentially, reaching about 2000W by 2030, and the peak power consumption of the rack will reach more than 300kW. On the one hand, improving power conversion efficiency is essential to reduce the operating costs of data centers; on the other hand, the large and rapidly fluctuating power curve on the rack side poses a challenge to the stability of public power grids, so power supply solutions are required to develop in the HVDC direction. SST, DRMOS, SiC, and GaN devices will become the core direction of AI power supplies. Focus on: Tianyue Advanced, Xinjie, Jianghai, Star Semiconductor, and Sunlink Electronics.

AI end side: AIAgent reshapes the interactive paradigm, big manufacturers scramble to lay out end side entrances, and the consumer electronics innovation year begins

As large models continue to evolve in multi-modal understanding, general reasoning, and task execution capabilities, AI is being upgraded from tool-based capabilities to AIagents that can understand user intentions and perform tasks independently. End-side consumer electronics products are a key carrier layer in the closed loop of AI commercialization, and are expected to systematically reconstruct the human-computer interaction paradigm. In this context, various terminal forms such as mobile phones, glasses, headsets, and home robots are expected to build collaborative networks around AIAgent to push AI from single-point functional upgrades to system-level experiences across scenarios and terminals. The importance of multi-modal inputs such as voice, vision, and environmental perception has increased, putting higher demands on end-side computing power, perception ability, and connectivity. The bank believes that the current end-side related technology and industry base has matured, and key breakthroughs in the business model are expected to form a “nonlinear amplification effect.” Looking ahead to 2026, from CES at the beginning of the year to WWDC in the middle of the year, or the continued exploration of leading manufacturers such as Byte and Apple, or the expectations of potential model makers and Internet vendors entering the market, it may become a key catalyst to ignite market sentiment and industry investment consensus. Focus on: Hengxuan Technology, Jingchen Co., Ltd., Lansi Technology, Lixun Precision, Lante Optics, Goertek, Aojie Technology, Lexin Technology, HuaQin Technology, Communication Holdings, Xiaomi Group, and Shunluo Electronics.

Semiconductors: An autonomous and controllable industrial chain where the recommendation process is expected to exceed expectations, and analog chips to accelerate domestic replacement during the economic recovery phase

According to SIA data, China accounted for 28% of global semiconductor sales in 2024, but the local supply ratio was only 4.5%. The self-sufficiency rate is still low, and since the increase mainly comes from cloud-side increments such as GPUs and HBM, the self-sufficiency rate is lower than in 2023. However, the financial performance of A-share semiconductor companies continued to improve. According to the operating data of 146 companies according to the bank's statistics, the highest quarterly revenue value fell to 54% in 2025. The overall gross margin of the 3Q25 SW semiconductor sector was between 1Q21 and 2Q21, and the net interest rate was comparable to 4Q20 and 1Q21. Looking at the industry cycle, global semiconductor sales have been growing year-on-year for eight consecutive quarters. In December '25, WSTS once again revised its forecasts for 2025 and 2026. It is expected that global semiconductors will achieve 3 consecutive double-digit growth in 2024-2026. Looking at the domestic semiconductor industry, in addition to the increase in AI, the rise of domestic chip design companies and local manufacturing demand are providing incremental growth for independent manufacturing chains, focusing on foundry, advanced packaging, and upstream semiconductor equipment materials; in addition, analog chips are lagging behind in the semiconductor product category, and the revenue of major international manufacturers TI and ADI began to correct year-on-year in 2025, marking that the industry has entered a stage of recovery. New products launched by domestic companies in recent years are expected to enter the scale release stage. In the long run, AI applications such as data centers and AI applications such as autonomous driving and humanoid robots have all brought them widespread use Incremental, and analog chips are also a segment with more room for localization, and will continue to benefit from increased localization rates. Concerns: SMIC, Huahong Semiconductor, Jewart, Sirup, Beifang Huachuang, China Micro, Tuojing Technology, Shengbang Co., Ltd., Nanxin Technology, Weitech Technology, Tongfu Microelectronics, Changdian Technology, Dinglong Co., Ltd., Jiaocheng Ultrasound.

Focus on portfolios: SMIC, IFF, China Micro, Aogee Technology, Huahong Semiconductor, Cambrian, Beifang Huachuang, Lanqi Technology, Lansi Technology, Lante Optics, Lixun Precision, Shenghong Technology, Huaqin Technology, Hengxuan Technology, Shunluo Electronics, Shanghai Electric Power Co., Ltd., Weitec Technology, Jingchen Co., Ltd., Pengding Holdings, Sirup, Jiehuart, Shengbang, Demingli, Jiang Bolong, Baiwei Storage, Goertek, Zhaoyi Innovation, Dinglong A. Changdian Technology, Crystal Optoelectronics, Communication Holdings, Haowei Group, Jiaocheng Ultrasound, Hikvision TV, Xiaomi Group, Tianyue Advanced, Yangjie Technology, Lexin Technology, Jianghai Co., Ltd., Electric Link Technology, Jingfeng Mingyuan, Guanghong Technology, Nanxin Technology, Shihua Technology, Beijing Junzheng, Jiemei Technology, Guoxin Technology, Tongfu Microelectronics, Weitou, and Fuliwang.

Risk warning: The domestic substitution process falls short of expectations; downstream demand falls short of expectations; risk of increased industry competition; risk of adverse changes in international relations; risk of cyclical industry fluctuations; risk of supply of production equipment and raw materials.