阿里集结AI铁三角“通云哥”，发布自研AI芯片“真武”对标英伟达H20

Core Insights - Alibaba has launched the "Zhenwu 810E" high-end AI chip, known as "Zhenwu" PPU, which was previously highlighted by CCTV [1][3] - The "Zhenwu" PPU features a self-developed parallel computing architecture and inter-chip interconnection technology, with a memory of 96G HBM2e and an interconnection bandwidth of 700 GB/s, suitable for AI training, inference, and autonomous driving [3] - The performance of "Zhenwu" PPU surpasses Nvidia's A800 and mainstream domestic GPUs, comparable to Nvidia's H20, and the upgraded version outperforms Nvidia's A100 [3] - The chip has been validated in external markets, receiving positive feedback for its stability and cost-effectiveness, with over 400 clients including State Grid, Chinese Academy of Sciences, and Xpeng Motors [3] Group 1 - The formation of the "Tongyun Ge" AI triangle, consisting of Tongyi Lab, Alibaba Cloud, and Pingtouge, marks a significant strategic move for Alibaba in the AI sector [4] - Alibaba has invested 17 years in strategic development, establishing a full-stack deployment of self-developed AI chips, a world-class cloud platform, and a leading large model [4] - The "Tongyun Ge" aims to create an AI supercomputer, leveraging self-developed chips, a leading cloud platform, and the world's strongest open-source model "Qianwen" for maximum efficiency [4] Group 2 - The recent release of the Qwen3-Max-Thinking model by Tongyi Lab has set multiple global records, rivaling GPT-5.2 and Gemini 3 Pro [5] - Alibaba Cloud's revenue growth has reached new highs, with AI-related product revenue achieving triple-digit year-on-year growth for nine consecutive quarters [5] - CEO Wu Yongming indicated a strong demand for AI, predicting a supply chain expansion cycle due to shortages in storage, CPU, and AI servers starting in the second half of 2025 [6] - Alibaba is accelerating its infrastructure investment, with a potential increase in the planned 380 billion yuan investment over three years, reflecting the growing AI demand [6]