Core Insights - TianShu ZhiXin (9903.HK), a leading domestic AI computing company, has announced a roadmap for its fourth-generation architecture that aims to match and surpass NVIDIA's architecture, with its "TongYang" edge products outperforming NVIDIA's AGX Orin [1][3] - The company has reported over 300 customers and 1,000 deployments, showcasing significant scaling achievements [1] Architecture Roadmap - The TianShu TianShu architecture, set to launch in 2025, is projected to exceed NVIDIA's Hopper architecture by 20% in computing power [1] - Subsequent architectures, TianShu TianXuan and TianShu TianJi, are expected to match and surpass NVIDIA's Blackwell in 2026, while the TianShu TianQuan architecture is anticipated to exceed NVIDIA's Rubin in 2027 [1] Product Performance - The "TongYang" series of edge products features measured dense computing power ranging from 100T to 300T, with the TY1000 outperforming NVIDIA's AGX Orin in computer vision and large model scenarios [3] - The TY1200 boasts an extreme performance of 300 TOPs, aiming to become the leading domestic edge computing solution [3]
天数智芯四代通用GPU架构 + 边端大算力产品齐发，国产算力商业化放量