1. Market Heat: Computing Power Sector Attracts 5.3 Billion in a Single Day, Cambricon Stock Price Returns to 1500 Yuan
On September 16, the semiconductor and computing power sector continued its strong performance, with Cambricon and Haiguang Information rising by 18% and 12% respectively, and Chunzong Technology’s liquid cooling server concept achieving five trading limits in six days. Funding data shows that institutional special seats net purchased 1.325 billion yuan of Zhongke Shuguang, and northbound funds increased their holdings in Cambricon by 870 million yuan.
Driving Factors:
-
Policy Support: The Ministry of Industry and Information Technology’s “Artificial Intelligence +” action plan clearly states that computing power investment will exceed 500 billion yuan by 2025.
-
Technological Breakthroughs: Huawei’s Ascend 910B chip performance is on par with NVIDIA’s A100, and the domestic GPU ecosystem is gradually improving.
-
Explosive Demand: The demand for computing power in AI large model training is increasing by 60% annually, and the global liquid cooling server market is expected to reach 12 billion USD by 2025.
2. Core Tracks and Investment Opportunities
1. Domestic GPUs: Breaking the Overseas Monopoly in the ‘Choke Point’ Field
-
Haiguang Information (688041): The successful tape-out of the Deep Computing No. 3 chip has improved performance by three times compared to the previous generation, and has received orders from Baidu’s Wenxin large model.
-
Cambricon (688256): The Siyuan 590 chip supports FP32 precision, with a 20% improvement in inference efficiency, and institutions predict revenue will exceed 5 billion yuan by 2025.
2. Liquid Cooling Servers: The ‘Cooling Revolution’ Under High Power Consumption
-
Zhongke Shuguang (603019): Immersion liquid cooling technology has a market share of over 60%, making it a core supplier for the “East Data West Computing” project.
-
Inspur (002837): The cold plate liquid cooling solution has been certified by Alibaba Cloud, with expected orders reaching 3 billion yuan by 2025.
3. AI Computing Power Chips: The ‘Heart’ of Computing Infrastructure
-
Cambricon Siyuan 590: Supports FP16 precision, with a 15% improvement in training efficiency compared to the A100, and has been deployed in SenseTime’s intelligent computing center.
-
Huawei Ascend 910B: Domestic production rate exceeds 90%, with training cluster computing power reaching 2.8 EFLOPS, and has been procured by ByteDance and Tencent Cloud.