Sinolink: Domestic computing power is accelerating throughout the full chain, with the potential for both quantity and price to rise together.

Sinolink released a research report stating that 2026 will be a crucial year for the transformation of China's computing power demand from "cloud training" to the dual-drive transformation of "training + inference". The computing power gap will be rapidly released under the catalysis of more modes and broader scenarios. The domestic computing power supply side will shift from a single scarcity state to structural balance, and abundant computing power resources will effectively meet the explosive demand, laying the foundation for the realization of the computing power industry chain's performance. Under the strong logic of supply and demand on both sides, the bank predicts that the computing power industry chain in 2026 will enter a "full-chain inflation" cycle, and industry prosperity will overflow from core chips to AIDC, cloud and computing services, supporting power equipment, and servers. Sinolink's main points are as follows: Cloud giants are raising prices, and major AI investments are increasing again. 1) Model transitions combined with Agent's deployment explosion will ignite the demand for inference, expanding the supply-demand gap for computing power, and the cost of computing infrastructure and explosive demand will be transmitted to the end, with AWS, Google Cloud, UCloud, Tencent Cloud, Aliyun, Baidu Intelligent Cloud and others successively announcing price adjustments for core products since the beginning of the year, indicating an unstoppable industry trend; 2) The commercial closed-loop of major AI companies is accelerating performance feedback driving continued investment in computing infrastructure. Alibaba's cloud business increased by 36% in FY26Q3, and it is expected that the annual revenue from cloud and AI commercialization, including MaaS, in the next five years, will exceed $100 billion, with Pinecone Brothers independently developed AI chips delivering 470,000 pieces as of February 2026; Tencent's core businesses like gaming, advertising, and social networking continue to grow under the influence of AI, with investment in AI product development amounting to 18 billion yuan last year and expected to double this year; Xiaomi's MiMo-V2-Pro has entered the top tier of large models in China, and announced plans to invest over 60 billion yuan in the field of AI in the next three years. The resonance of training and inference releases computing power demand rapidly. The bank predicts that 2026 will be a key year for the dual-drive transformation of China's computing power demand from "cloud training" to "training + inference", and the computing power gap will be rapidly released under the catalysis of more modes and broader scenarios. 1) Training side: advancing towards high quality and multi-modal. Leading internet companies (ByteDance, Alibaba, Tencent) continue to iterate trillion-parameter models, with new forces such as KNOWLEDGE ATLAS and DeepSeek quickly updating the MoE architecture. The scaling-law in the multi-modal field continues significantly, with models like Seedance promoting a unified understanding of video, audio, and text, driving bottom-level computing power requirements from single text to high-consumption video/3D training, which requires higher demands for cluster interconnection bandwidth and stability. 2) Inference side: the year of application landing, with steep demand slope. In 2026, AI applications will accelerate penetration, according to QuestMobile data, Douyin's app MAU exceeded 226 million in December 2025, and with the added exposure effect of the Spring Festival Gala, activity is expected to reach new heights; TONG YI QIAN WEN has also fully opened up the Alibaba ecosystem. C-end traffic and AI animated dramas, programming, and other native scenes explode, coupled with the B-end vertical model's momentum, collectively driving a substantial increase in real-time inference computing power consumption. The bank predicts that inference-side demand will become a new engine driving industry chain growth. External marginal improvement on the supply side, internal localization volume. The bank believes that in 2026, China's domestic computing power supply side will shift from a single scarcity state to structural balance, and abundant computing power resources will effectively meet the explosive demand, laying the foundation for the realization of the computing power industry chain's performance. 1) Externally, according to the update of relevant trade license lists, NVIDIA's H200 (compliant version) has been officially approved to enter the Chinese market, which will effectively alleviate the computing power anxiety of major internet companies in ultra-large-scale model training in the short term and accelerate the model iteration speed. 2) At the same time, the performance and ecological development of domestic computing power chips have crossed the turning point from "available" to "usable". Huawei's Ascend series, Cambricon's MLU series, and Horizon Robotics' deep learning series have rapidly iterated in practical applications; major companies' self-developed chip strategies are entering the harvest period, with Baidu Kunlun, Alibaba's Pinecone Brothers, and ByteDance's self-developed chips starting large-scale deployment; and startups like Moore Thread, Mu Xi, ILUVATAR COREX are also accelerating the IPO process, further enriching the market supply. 3) The expansion of upstream advanced process capacity provides a solid foundation for chip supply. Semiconductor Manufacturing International Corporation's full-year financial report for 2025 just released shows breakthrough progress in expanding production in advanced and mature processes, with steadily increasing capacity utilization throughout the year and high capital expenditure to ensure timely release of new capacity. Domestic computing power industry chain prosperity accelerates, with the possibility of jointly rising in quantity and price. Under the strong logic of supply and demand on both sides, the bank predicts that the computing power industry chain in 2026 will enter a "full-chain inflation" cycle, and industry prosperity will overflow from core chips to AIDC, cloud and computing services, supporting power equipment, and servers. In terms of investment strategy, considering the relatively clear capital expenditure trends of giants such as ByteDance, Alibaba, and Tencent, their supply chain has a higher level of performance certainty, so deeply linking supply chain companies with major internet companies (CSP) may achieve significant excess returns. Related targets: Guangdong Hec Technology Holding, Cambricon, Hygon Information Technology, Range Intelligent Computing Technology Group, Jiangsu Lettall Electronic, Henan Yuneng Holdings, Sharetronic Data Technology, UCloud, Sichuan Huafeng Technology, Wangsu Science & Technology, Digital China Group, Cloud Light, Dawei Technology (Guangdong) Group, RunJian Co., Ltd., Zhejiang Entive Smart Kitchen Appliance, Kehua Data Co., Ltd., Semiconductor Manufacturing International Corporation, HUA HONG SEMI, Dawning Information Industry, Suzhou Hesheng Special Material, Guangdong Aofei Data Technology, Capitalonline Data Service, INESA Intelligent Tech Inc., Zhejiang Risun Intelligent Technology, Inspur Electronic Information Industry, Weichai Heavy Machinery, Shenzhen Honor Electronic, etc. Risk warnings: Risks of intensified industry competition; risks of slower-than-expected progress in technological research and development; risks of cyclical fluctuations in capital expenditures in specific industries.