手握原创GPU架构与12.4亿元订单，壁仞科技赴港上市受热捧

Core Viewpoint - Wallan Technology has initiated a global offering through the Hong Kong Stock Exchange, attracting 23 cornerstone investors, including top-tier institutions, with a cornerstone subscription ratio of approximately 64% based on the median issue price. The company is positioned as a significant competitor in the domestic GPGPU market, often referred to as the "first GPU stock in Hong Kong" [1]. Financial Performance - Wallan Technology's revenue is projected to increase significantly from RMB 0.5 million in 2022 to RMB 620.3 million in 2023, and further to RMB 3.37 billion in 2024 [2]. - The company reported losses of RMB 1.474 billion, RMB 1.744 billion, RMB 1.538 billion, and RMB 1.6 billion for the years 2022, 2023, 2024, and the first half of 2025, respectively. Adjusted net losses are expected to narrow over the same period [3][4]. Research and Development - Wallan Technology's R&D expenditures reached RMB 10.18 billion, RMB 8.86 billion, RMB 8.27 billion, and RMB 5.72 billion from 2022 to the first half of 2025, accounting for 79.8%, 76.4%, 73.7%, and 79.1% of total operating expenses, respectively [6]. - The company has developed its first-generation GPGPU architecture and two chips, BR106 and BR110, and is advancing to the BR166 chip, which utilizes advanced packaging technologies [6][7]. Market Position and Orders - As of December 15, 2025, Wallan Technology has secured orders worth approximately RMB 12.4 billion, indicating strong commercial prospects. The company has established strategic partnerships with key clients in high-demand sectors such as AI data centers, telecommunications, and fintech [8]. - Wallan Technology has received significant backing from notable investors, including state-owned platforms and top-tier venture capital firms, enhancing its commercial viability [8]. Product Development and Future Outlook - The company plans to launch the next-generation BR20X series data center chips by 2026, which will enhance computational capabilities and support a wider range of data formats [10]. - Future products, including the BR30X and BR31X, are in the planning stages, expected to be commercialized by 2028, focusing on cloud training and edge inference [11].