A Strategic Shift Towards Domestic AI Hardware
Chinese technology giants Alibaba and Baidu have begun integrating their own internally designed chips for training artificial intelligence models, marking a significant step towards reducing their reliance on foreign hardware, particularly from U.S. chipmaker Nvidia. This move aligns with China's broader national objective of achieving self-sufficiency in critical technologies, a push intensified by increasing U.S. export controls on advanced semiconductors.
Alibaba's Growing Chip Portfolio
Alibaba's semiconductor division has developed the Zhenwu processing unit, which the company has been utilizing for training smaller AI models since early 2025. While Alibaba previously launched the Hanguang 800 AI-inference chip in 2019, designed for specific applications like image search and smart city solutions, the newer Zhenwu chip is engineered for a broader range of AI inference tasks. A notable development is the shift in manufacturing, with Alibaba's latest AI chip now being produced by a Chinese company, departing from earlier processors fabricated by Taiwan Semiconductor Manufacturing Company (TSMC). Reports indicate that Alibaba's current AI chip capabilities are becoming competitive with Nvidia's H20, the most powerful AI processor Nvidia is permitted to sell in China.
Baidu's Kunlun Series Advances AI Training
Baidu has been a pioneer in developing its own AI accelerators under the Kunlun series. The company first unveiled its Kunlun 818-300 model chip in 2018, optimized for various AI tasks including voice recognition, natural language processing, and autonomous driving. By August 2021, Baidu's Kunlun II processor entered volume production, designed for cloud, edge, and autonomous vehicle applications. Currently, Baidu is leveraging its latest Kunlun P800 chip to train newer versions of its flagship Ernie AI model. The Kunlun chips are positioned by Baidu as both a cost-saving measure and a strategic hedge against potential foreign sanctions.
Implications Amidst U.S. Export Restrictions
The increasing adoption of in-house chips by Alibaba and Baidu is a direct response to the progressively tightened U.S. export controls on advanced AI chips to China, which began in October 2022 and were expanded in 2023 and 2024. These restrictions aim to limit China's access to high-end chips crucial for AI development and supercomputing. While Chinese tech companies have historically relied heavily on Nvidia's powerful processors, the U.S. measures have spurred Beijing's push for a domestic semiconductor ecosystem. Although neither Alibaba nor Baidu has completely abandoned Nvidia hardware, continuing to use it for their most cutting-edge models, the shift towards homegrown alternatives signifies a growing challenge to Nvidia's market share in China. This trend underscores China's determination to build a resilient, self-sufficient AI industry.
0 Comments