Under the U.S. chip export controls, Chinese tech giants are accelerating efforts to fill the market gap left by NVIDIA. Alibaba’s newly launched AI chip not only challenges NVIDIA’s market dominance but also reveals China’s strategic determination for self-sufficiency in the AI computing power supply chain.
On August 29, 2025, Alibaba announced the successful development of a new AI inference chip, which is currently in the testing phase and aims to handle a broader range of artificial intelligence inference tasks while being compatible with NVIDIA architecture.
Unlike previous chips that relied on TSMC for manufacturing, this new chip will be produced by domestic companies in mainland China. This shift marks a significant step for China in enhancing its chip self-production capabilities, aiming to reduce dependence on overseas supply chains.
01 International Situation and Policy Background
The U.S. government’s export restrictions on chips to China have created a market vacuum for AI chips. Earlier this year, NVIDIA’s H20 chip, which was allowed to be sold in China, faced a ban from the U.S., although it was reinstated in July. However, Chinese companies have begun to continuously develop alternative products.
Beijing is supporting this push through significant investments, including an $8.4 billion AI fund announced in January. The year 2025 marks the conclusion of China’s 14th Five-Year Plan for semiconductors, making it urgent for domestic chips to be realized.
Recently, the U.S. has tightened exports of lithography machines, forcing domestic companies to rely more on local supply chains. While domestic chips can currently handle AI inference tasks, high-end chips for training large models still need to be imported.
02 Technical Breakthroughs of Alibaba’s AI Chip
Alibaba’s new chip primarily targets AI inference tasks, which involve applying pre-trained models rather than training the models themselves.
The chip’s compatibility with NVIDIA architecture means that engineers can reuse programs they have written for NVIDIA chips, significantly reducing the migration costs for developers. This compatibility design is believed to allow Chinese AI developers to “close the gap with the U.S. faster than most people imagine.”
In addition to Alibaba, the Shanghai startup MetaX also launched a chip in July that claims to be a substitute for the H20, offering larger memory but higher power consumption. Huawei has demonstrated a computing system integrating 384 Ascend chips, with some analysts stating that despite higher power consumption, it is more powerful than NVIDIA’s top systems in certain metrics.
## 03 Analysis of Beneficiary Sectors and Leading Stocks in the Industry Chain
Alibaba announced plans to invest 380 billion yuan over the next three years in cloud and AI hardware infrastructure, focusing on areas such as chips, computing power infrastructure, liquid cooling technology, and industry applications.
Chip Design and Manufacturing
Cambricon (688256): Core supplier of Alibaba Cloud’s AI inference chip, the Siyuan 590 chip is compatible with Alibaba Cloud servers, a key target for domestic substitution, receiving an initial order of 1 billion yuan from Alibaba in Q2 2025.
Haiguang Information (688041): The DCU series GPU is compatible with the CUDA ecosystem, clearly replacing the demand for NVIDIA A100, with AI chip revenue expected to grow by 287% in 2024, and jointly developing the next-generation training card with Alibaba.
SMIC (688981): As a leading domestic wafer foundry, its 7nm/N+1 process may gain new order increments.
Computing Power Infrastructure and Servers
Data Harbor (603881): Core supplier of customized IDC for Alibaba Cloud, constructing over 20 data centers, accounting for more than 80% of revenue.
Inspur Information (000977): Main supplier of Alibaba AI servers, with liquid cooling technology reducing PUE to below 1.1, expected to receive over 50% of new cabinet orders from Alibaba.
– **Hangang Co. (600126): Provides 30% of inference computing power for Alibaba Tongyi large model, receiving an order of 15 billion yuan in Q1 2025, with liquid cooling technology PUE as low as 1.15.
Liquid Cooling Technology
Inspur (002837): As a leading company in the liquid cooling technology field, it will directly benefit from the promotion of Alibaba’s AI chips, holding a 40% share in Alibaba’s data center liquid cooling procurement.
Shenling Environment (301018): In the first half of 2025, the company’s data center temperature control business achieved revenue of 1.25 billion yuan, a year-on-year increase of 65%, with liquid cooling products accounting for 35% of the total.
Ecological Application Cooperation
Hang Seng Electronics (600570): Collaborating with Alibaba Cloud to build financial AI solutions covering intelligent investment advisory and risk control.
Qianfang Technology (002373): Partnering with Alibaba Cloud on “City Brain 2.0,” winning over 800 million yuan in Alibaba ecological projects in H1 2025.
Runhe Software (300339): Developing edge computing modules based on Alibaba’s Xuantie processor, with Cainiao Logistics holding over 30% market share.
04 Risks and Opportunities from an Investment Perspective
Short-term market impact: Alibaba’s AI chip is expected to quickly open up the market in the short term, alleviating the “computing power shortage” faced by domestic AI companies. This chip supports the CUDA architecture and is compatible with the NVIDIA ecosystem, significantly lowering the barriers to replacement.
Supply chain self-control: The production model relying on domestic companies completely eliminates dependence on TSMC, achieving supply chain self-control against the backdrop of increasing geopolitical risks.
Performance benchmarking against international standards: The chip, currently in the testing phase, has computing power close to NVIDIA A10 and supports FP8 precision, showing good adaptability for multi-modal large model inference.
Ecological synergy effect: This chip is perfectly compatible with Alibaba Cloud’s machine learning platform (PAI), closely integrating with applications such as Taobao and Alipay, forming a “chip-cloud-end” closed-loop ecosystem.
Challenges and risks: Surveys show that 47% of companies are still observing, concerned about the long-term stability of domestic chips. There is uncertainty in technology transfer, and the liquid cooling business is still in the investment phase, requiring verification of cooling efficiency and cost control. There are also risks of order dependency, with revenues highly reliant on the mass production progress of Alibaba’s chips.