财联社·2025-09-02 00:34Core Viewpoint - Alibaba Cloud is facing a potential computing power shortage for its Tongyi Qianwen large model, prompting the company to reportedly increase its order of Cambricon's Siyuan 370 chips to 150,000 units, although this claim has been denied by Alibaba Cloud representatives [2][3]. Group 1: AI Chip Market Dynamics - Major players in the domestic AI chip market include Cambricon, Huawei, Haiguang Information, Birun, Muxi, Suyuan, and Moore Threads [4]. - According to IDC, the market scale for accelerated chips in China is expected to exceed 2.7 million units by 2024, with GPU cards holding a 70% market share [4]. - Domestic AI chip brands have shipped over 820,000 units, with Huawei's Ascend series capturing a significant portion of the market [4]. Group 2: Alibaba's Chip Development Strategy - Alibaba is actively developing its own chips through its independent semiconductor company, Pingtouge, established in 2018 [6]. - Pingtouge has launched several chip series, including the "Xuantie" RISC-V processors and "Hanguang" AI chips, with some chips already deployed at scale on Alibaba Cloud [6]. - Reports suggest that Alibaba's new AI chip, which is compatible with Nvidia, is currently in testing and will be manufactured by a domestic company instead of TSMC [6][7]. Group 3: Competitive Landscape in AI Chips - Other internet companies like Baidu, ByteDance, and Tencent are also exploring chip development [8]. - Baidu's Kunlun chip supernode has been fully operational since August, supporting extensive AI model training [8]. - Tencent has introduced several self-developed chips, including AI inference and video transcoding chips, and has collaborated with AMD on GPU cards [8]. Group 4: Strategic Importance of Chip Supply Chains - Establishing a self-sufficient supply chain that includes domestic chips is crucial for the future development of the AI ecosystem [8]. - Alibaba's "One Cloud, Multiple Chips" strategy aims to ensure compatibility with various chip architectures, including X86, ARM, and RISC-V [8]. - Experts emphasize the need for domestic chips to enhance performance and build ecosystems to meet global computing power challenges [9].