According to reports, China is intensifying its investments in the artificial intelligence (AI) sector. UBS has released a research note indicating that despite uncertainties surrounding the import of AI chips, domestic computing capabilities continue to evolve, backed by national policy support and significant investments in R&D by major tech firms and local suppliers. This trend is expected to further propel the development of AI and large models in China.
The firm is optimistic about Alibaba (BABA-W) and Baidu (BIDU-SW), believing that advancements in self-developed chips will strengthen their positions in the AI value chain while continuing their investments in AI. UBS highlighted recent positive technological factors, including: 1) the rapid improvements in performance despite existing gaps in chip technology, driven by ongoing R&D efforts in Chinese internet companies and local GPU suppliers; 2) system-level enhancements through supernode scaling, such as Alibaba's Panjiu 128 supernode and Huawei's Ascend 384 supernode, which significantly increase the number of GPUs per rack and help mitigate performance gaps seen in individual domestic GPUs, thus achieving higher rack-level computing power. They believe these designs enable domestic chips to support more complex reasoning scenarios. In the long run, network technology advancements could lead to the scaling out of nodes to large clusters, potentially supporting training workloads. 3) AI model developers are optimizing algorithms specifically for domestic GPUs. The latest DeepSeek v3.2 model utilizes the domestic GPU programming language TileLang, which better adapts to the local algorithmic ecosystem like Huawei Ascend and Cambricon (688256.SH).
UBS noted that most internet companies are accelerating the development of application-specific integrated circuits (ASICs) to optimize their workloads and enhance cost-effectiveness. Google was among the first to develop its own AI chips, having gone through multiple iterations, and other companies like Amazon, META, and Microsoft are following suit with their customized AI chips. In China, Baidu has developed three generations of Kunlun chips, and Alibaba has also begun the deployment of its self-developed chips.
Following a recent survey of AI chip experts, UBS summarized three key points: 1) Hardware performance: The computing power of leading domestic GPUs now matches NVIDIA’s (NVDA) Ampere architecture, with next-generation products targeting Hopper, but still lagging a generation behind the Blackwell series. 2) Software ecosystem: Some domestic chip manufacturers have established their own software stacks or added CUDA compatibility through translation tools, enhancing engineers’ migration efficiency; however, the fragmentation of the ecosystem limits scalability. 3) Supply chain capabilities: Apart from chip design quality, China's capabilities in advanced process technology and high-bandwidth memory production remain in early stages.
In addition to Alibaba and Baidu, UBS is also optimistic about iFlytek (002230.SZ) due to its unique positioning and lead in integrating domestic hardware with large model development. The firm also favors Horizon Robotics (09660), Northern Huachuang (002371.SZ), and Zhongwei Company (688012.SH).