Guojin Securities: Domestic Computing Power Full-Chain Sentiment Accelerating, Expected Volume and Price Increases

CITIC Securities APP has learned that Guojin Securities released a research report stating that 2026 will be a critical year for China’s computing power demand to shift from “cloud training” to a dual-driven model of “training + inference.” The computing gap will rapidly release under the catalysis of more modalities and broader scenarios. Domestic computing supply will shift from a single shortage to structural balance, with abundant resources effectively supporting explosive demand, laying a foundation for performance realization in the computing industry chain. Under strong supply and demand logic, it is forecasted that in 2026, the computing industry chain will enter a “full-chain inflation” cycle, with industry prosperity spilling over from core chips to AIDC, cloud and computing services, supporting power equipment, servers, and other segments.

Guojin Securities’ main points are as follows:

Cloud providers raise prices, major companies increase AI investments again. 1) Model breakthroughs combined with Agent breakthroughs ignite inference demand, expanding the supply-demand gap for computing power. The cost of infrastructure and explosive demand are transmitted to end users. Since the beginning of the year, AWS, Google Cloud, UCloud, Tencent Cloud, Alibaba Cloud, Baidu Smart Cloud, etc., have announced core product price adjustments, indicating an unstoppable industry trend; 2) Major companies accelerate the realization of AI commercial cycles, with performance feedback driving continued investment in infrastructure. Alibaba’s FY26Q3 cloud business grew by 36%, and it is expected that in the next five years, cloud and AI commercialization including MaaS will exceed $100 billion annually. Pingtouge’s self-developed AI chips have delivered 470,000 units by February 2026; Tencent’s core businesses such as gaming, advertising, and social media continue to grow with AI support, with at least doubling of investment this year after investing 18 billion yuan last year in new AI products; Xiaomi’s MiMo-V2-Pro has entered the top tier of domestic large models and announced plans to invest over 60 billion yuan in AI over the next three years.

Training and inference resonance, rapid release of computing demand. The report predicts that 2026 will be a key year for China’s computing demand to shift from “cloud training” to a dual-driven model of “training + inference,” with the gap rapidly releasing under more modalities and broader scenarios. 1) On the training side: advancing towards high quality and multi-modality. Leading internet companies (ByteDance, Alibaba, Tencent) continue iterating trillion-parameter models, while new forces like Zhipu and DeepSeek rapidly update MoE architectures. Scaling laws continue to extend in multi-modality, with models like Seedance achieving unified understanding of video, audio, and text, pushing underlying computing needs from single text to high-consuming video/3D training, demanding higher cluster interconnect bandwidth and stability. 2) On the inference side: application deployment year, demand growth steep. AI applications accelerate in penetration; according to QuestMobile data, Doubao APP’s MAU exceeded 226 million in December 2025, with exposure from the Spring Festival Gala potentially boosting active users further; Tongyi Qianwen fully integrates into Alibaba’s ecosystem. Consumer traffic and native scenarios like AI dramas and programming are exploding, combined with B-end vertical models, jointly driving significant growth in real-time inference computing power. The forecast is that inference demand will become a new engine for industry chain growth.

External marginal improvement in supply, domestic production volume increases. The report believes that by 2026, domestic computing supply will shift from a single shortage to structural balance, with abundant resources supporting explosive demand and underpinning industry performance. 1) Externally, based on updates to trade permit lists, NVIDIA H200 (compliant version) has officially been approved for entry into China, which will alleviate the computing anxiety of leading internet companies in training large-scale models and accelerate model iteration. 2) Meanwhile, domestic chips have crossed the “usable” to “good” threshold in performance and ecosystem development. Huawei’s Ascend series, Cambrian’s Syuanyuan series, and Hygon’s DeepSea series are rapidly iterating in practice; major companies’ self-developed chips are entering a harvest period, with Baidu’s Kunlun chips, Alibaba’s Pingtouge, and ByteDance’s self-developed chips beginning large-scale deployment; startups like Moore Thread, MuXi, and Tiannanshi are accelerating IPO processes, further enriching market supply. 3) Expansion of upstream advanced process capacity provides a fundamental guarantee for chip supply. SMIC’s 2025 annual financial report shows breakthroughs in both advanced and mature process capacity expansion, with steadily rising utilization rates and high capital expenditure to ensure timely release of new capacity.

Full-chain domestic computing power accelerates, with potential for volume and price increases. Under the strong logic of supply and demand, it is forecasted that in 2026, the computing industry chain will enter a “full-chain inflation” cycle, with industry prosperity spilling over from core chips to AIDC, cloud and computing services, supporting power equipment, servers, and other segments. Investment strategies suggest that, given the clear capital expenditure trends of giants like ByteDance, Alibaba, and Tencent, their supply chains have high performance certainty. Companies deeply linked to top internet CSPs’ supply chains may achieve significant excess returns.

Related stocks: Dongyangguang, Cambrian, Hygon Information, Runze Technology, Litong Electronics, Yunnan Energy Holdings, Xiechuang Data, UCloud, Huafeng Technology, Wangsu Technology, Digital China, Yuntian Lifei, Dawi Technology, Runjian Shares, Yitian Smart, Kehua Data, SMIC, Huahong Semiconductor, Sugon, Inspur, Esense New Materials, Aofei Data, Capital Online, Yunsai Zhijian, Ruisheng Intelligent, Inspur Information, Weichai Heavy Machinery, Oulu Tong, etc.

Risk warning: Risks of intensified industry competition; risks of slower-than-expected technological R&D progress; risks of cyclical fluctuations in downstream capital expenditure in specific industries.