天数智芯发布四代架构路线图：预期于2027年超越英伟达Rubin架构

Core Viewpoint - TianShu ZhiXin (stock code: 09903.HK) has released a roadmap for its fourth-generation architecture, aiming to achieve "high-quality computing power" with a focus on efficiency, predictability, and sustainability, and plans to surpass NVIDIA's Rubin architecture by 2027 [2][5]. Group 1: Fourth-Generation Architecture - The fourth-generation architecture includes several key designs: TianShu TianShu architecture will surpass Hopper by 2025, TianShu TianXuan architecture will match Blackwell by 2026, and TianShu TianJing architecture will exceed Blackwell by 2026, with TianShu TianQuan architecture expected to surpass Rubin by 2027 [4][5]. - The architecture aims to address industry challenges such as low energy efficiency and insufficient creativity by defining high-quality computing power with three core characteristics: high efficiency, predictability, and sustainability [7]. Group 2: AI++ Computing System - The AI++ computing system paradigm integrates both chip-level and external computing systems, establishing a new model driven by software to unlock hardware potential and support a thriving application ecosystem [9]. Group 3: "Tongyang" Series Product Launch - The "Tongyang" series of edge computing products was launched, featuring four models designed for various applications, with performance metrics exceeding NVIDIA's AGX Orin [14][16]. - The series aims to provide high computing power for physical AI applications, with performance ranging from 100T to 300T in various scenarios [16]. Group 4: Industry Applications and Value - TianShu ZhiXin has demonstrated the successful application of its products in multiple sectors, including internet, finance, healthcare, and research, showcasing the maturity and reliability of domestic computing power [19]. - The company has achieved significant performance improvements across various industries, such as doubling single-machine performance in AI and reducing token costs by 50% in the internet sector [21]. Group 5: Ecosystem Development - The company is enhancing the domestic AI computing ecosystem by signing strategic cooperation agreements with hardware manufacturers and solution providers, aiming to create an open and win-win industrial ecosystem [27]. - The CEO emphasizes the importance of a fully self-developed stack to solidify the ecological foundation and promote the prosperity of the domestic computing ecosystem [29].