Alibaba Advances AI Chip Development Amid U.S. Export Restrictions

Alibaba Group Holding Limited is enhancing its AI infrastructure capabilities by introducing the Zhenwu 810E, a sophisticated AI chip developed entirely in-house by its semiconductor design arm, T-Head. This strategic initiative responds to growing restrictions on China’s access to advanced semiconductors, imposed through U.S. export controls, which have complicated procurement of foreign chips such as those from Nvidia Corporation. T-Head publicized the technical specifications of the Zhenwu 810E, describing it as an application-specific integrated circuit (ASIC) built to efficiently manage both AI training and inference tasks. The chip’s design targets the high data throughput demands intrinsic to generative AI models and offers performance comparable to Nvidia's H20 GPU, which has been adapted for the Chinese market. This is significant given that prior to the escalation of U.S.-China regulatory tensions, Nvidia reportedly controlled approximately 95% of China’s advanced AI chip market. Alibaba has already integrated the Zhenwu 810E chip at scale within its cloud computing infrastructure, deploying it across a multitude of clusters, each comprising tens of thousands of computing cards. Beyond Alibaba’s own ecosystem, the chip has attracted over 400 external customers, including prominent organizations such as State Grid, the Chinese Academy of Sciences, and the electric vehicle manufacturer XPeng. Additionally, the chip has been optimized to support the training and inference of Alibaba’s proprietary Qwen large language models, suggesting close alignment with the company’s AI service capabilities. Meanwhile, market data indicates that despite evolving supply dynamics, Nvidia continues to ship substantial volumes of H200 GPUs—an estimated 400,000 units—to major Chinese technology firms including ByteDance, Alibaba itself, and Tencent Holdings. This demonstrates the ongoing interplay between dependence on foreign AI hardware and domestic chip development efforts. Amid this backdrop, Alibaba and peer Chinese technology giant Baidu are progressing plans to spin off their semiconductor units. Alibaba intends to restructure T-Head into a partially employee-owned enterprise with the prospect of a public listing in the future. Founded in 2018, T-Head specializes in designing computing and storage chips critical for AI and cloud infrastructure. Concurrently, Baidu is preparing an initial public offering for its Kunlunxin chip division on the Hong Kong stock exchange, targeting a capital raise of up to $2 billion while retaining a majority ownership stake of approximately 59%. In terms of market reaction, Alibaba’s stock closed lower in premarket trading on Friday, fitting a broader pattern of volatility linked to regulatory tensions and competitive developments. Shares slipped around 1.24% to approximately $172.09 per share. This tangible shift toward proprietary AI hardware by Alibaba reflects a broader strategic focus on reducing exposure to external supply chain risks, particularly those tied to geopolitical constraints. It also underscores the increasing importance placed on AI chip performance and scalability to support next-generation AI applications within China’s large digital economy. Investors and industry watchers will closely monitor the performance and adoption of the Zhenwu 810E chip across Alibaba’s cloud services and external clients, as well as any progress on the anticipated public listing of T-Head. The unfolding developments in AI semiconductor technology and market structure are critical to understanding Alibaba’s positioning in the evolving AI landscape.