Core Insights - Alibaba's chip company, PingTouGe, has launched a high-end AI chip named "Zhenwu 810E," marking a significant step in its AI strategy and the formation of the "Tongyun Ge" AI supercomputer with Alibaba Cloud and Tongyi Lab [1][2] - The "Zhenwu" PPU has been deployed in multiple clusters on Alibaba Cloud, serving over 400 clients, showcasing PingTouGe's capabilities in the chip sector [2] - Alibaba and Google are currently the only two companies globally with top-tier capabilities across large models, cloud, and chip technologies [3] Group 1 - The "Zhenwu" PPU features a self-developed parallel computing architecture and inter-chip communication technology, with 96G HBM2e memory and a bandwidth of 700 GB/s, suitable for AI training, inference, and autonomous driving [7] - The performance of the "Zhenwu" PPU surpasses Nvidia's A800 and mainstream domestic GPUs, comparable to Nvidia's H20, and the upgraded version outperforms Nvidia's A100 [7] - The launch of the "Zhenwu" PPU is part of Alibaba's 17-year strategic investment and vertical integration in AI, which includes Alibaba Cloud, PingTouGe, and the Qianwen model [2][7] Group 2 - The chip concept is closely tied to Alibaba's market valuation, with recent news of PingTouGe's potential independent listing leading to a market capitalization increase of approximately HKD 100 billion in two days [8] - The Qianwen flagship reasoning model, Qwen3-Max-Thinking, has set multiple global records, demonstrating performance comparable to GPT-5.2 and Gemini 3 Pro [7] - The Qianwen open-source model has seen over 200,000 derivative models and more than 1 billion downloads, maintaining its position as the largest in the world [7]

Group 1 - The core point of the article is the launch of Alibaba's high-end AI chip "Zhenwu 810E" and the formation of the AI supercomputer "Tongyun Ge," which integrates self-developed chips, cloud services, and open-source models [1][3] - Alibaba aims to create an AI supercomputer that leverages its self-developed chip, the leading cloud service in the Asia-Pacific region, and the world's strongest open-source model "Qianwen" for maximum efficiency in training and deploying large models [3] - The "Zhenwu" PPU has been deployed in multiple clusters on Alibaba Cloud, serving over 400 clients, including State Grid, Chinese Academy of Sciences, Xiaopeng Motors, and Sina Weibo [3] Group 2 - The "Zhenwu" PPU features a self-developed parallel computing architecture and inter-chip communication technology, with a memory of 96G HBM2e and an inter-chip bandwidth of 700 GB/s, suitable for AI training, inference, and autonomous driving [3] - The performance of the "Zhenwu" PPU surpasses that of Nvidia's A800 and mainstream domestic GPUs, being comparable to Nvidia's H20 [3]

【#阿里自研高端AI芯片曝光# 黄金三角"通云哥"浮出水面】继传出预计"独立上市"之后，阿里旗下芯 片公司"平头哥"日前再传大消息。 2026年1月29日上午，平头哥官网悄然上线一款名为"真武810E"的高端AI芯片，此前被央视《新闻联 播》曝光的阿里自研芯片PPU正式亮相。行业资深人士认为，这也意味着通义实验室、阿里云和平头哥 组成的阿里巴巴AI黄金三角"通云哥"首次浮出水面。 《中国经营报》获悉，阿里巴巴正在将"通云哥"打造成一台AI超级计算机，它同时拥有全栈自研芯片平 头哥、亚太第一的阿里云，以及全球最强的开源模型"千问"，可以在芯片架构、云平台架构和模型架构 上协同创新，从而实现用户在阿里云上训练和调用大模型时达到最高效率。 据悉，"真武"PPU已在阿里云实现多个万卡集群部署，服务国家电网、中科院、小鹏汽车、新浪微博等 400多家客户。 "真武"PPU的正式亮相，显示了平头哥在芯片领域积累多年的实力。阿里巴巴2009年创建阿里云，2018 年成立平头哥芯片公司，2019年启动大模型研究，经过长达17年的战略投入和垂直整合，终于实现"通 云哥"全栈AI的完整布局。 阿里巴巴长期观察人士认为，目前整个阿里 ...

Core Insights - Alibaba has confirmed the existence of its self-developed AI chip, PPU, named Zhenwu 810E, which utilizes GPGPU technology and is designed for AI training, inference, and autonomous driving applications [1][3] - The PPU chip has been in development since 2020 and has recently begun commercial sales to various domestic computing service providers and server manufacturers [2][3] - As of early 2026, the PPU chip has been widely deployed, serving over 400 clients, including major organizations like State Grid and Xpeng Motors [3] Group 1: Chip Development and Technology - The Zhenwu 810E chip features 96G HBM2e memory and an interconnect bandwidth of 700GB/s, positioning it competitively against NVIDIA's GPUs [1] - The chip's development was kept internal until its recent unveiling, indicating a strategic approach to market entry [1][2] - Internal evaluations suggest that the Zhenwu 810E outperforms NVIDIA's A800 and is comparable to the H20 model [3] Group 2: Market Position and Competition - By early 2025, Alibaba began actively selling the PPU chip, indicating a shift towards commercialization [2] - In the first half of 2025, Alibaba's market share in the domestic AI chip market ranked second, following Huawei's Ascend series [3] - The emergence of multiple AI chip companies with significant shipment volumes indicates a growing competitive landscape in the domestic market [4][5] Group 3: Industry Trends and Future Outlook - The increasing shipment volumes of AI chips, with at least nine companies exceeding 10,000 units, reflect a maturation of the domestic AI chip industry [4][5] - The price range for domestic AI chips is between 30,000 to 200,000 yuan per unit, suggesting a market acceptance of their performance and stability [5] - Industry experts anticipate a surge in the shipment of domestic AI inference chips as manufacturing capacities improve in 2026 [5][6]

1月29日，记者关注到，阿里平头哥官网上线了自研AI芯片"真武810E"的产品信息，这也是2025年9月《新闻联播》披露的平头哥自研芯片"PPU"的正式亮 相。 对于镇岳510，平头哥早早做下了"要完全跑通产业化、商业化"的产品定位，2023年至2025年，镇岳510用一年多的时间，在阿里云内部获得了规模化的、充 分的验证，证明除了云服务、AI外，其顶尖性能对高性能企业存储（AI、OLTP等）也有很大的吸引力。 目前，镇岳510已在阿里云EBS规模化部署，广泛应用于AI训练、AI推理、分布式存储、在线交易等业务场景。同时，镇岳510与忆恒创源、得瑞领新、佰维 存储、长江万润等存储厂商的合作及生态建设也在加快。 此外，平头哥发布的公开信息显示，2025年9月，平头哥羽阵系列产品的规模化应用也进入了新阶段。2025 IOTE国际物联网展上，平头哥展出了羽阵611超 高频RFID电子标签芯片在零售、物流、供应链、洗涤和畜牧管理等行业的解决方案，并首次公开了与菜鸟联合打造的RFID解决方案，麦当劳中国成为这个 解决方案的大客户之一。 根据平头哥官网介绍，"真武"PPU采用自研并行计算架构和片间互联技术，配合全栈自研软 ...

Core Viewpoint - Alibaba is establishing an AI supercomputer named "Tongyun Ge," integrating its self-developed chips, cloud services, and advanced models to enhance efficiency in training and deploying large models [2][4]. Group 1: AI Chip Development - The "Zhenwu 810E" high-end AI chip has been launched by Alibaba's Pingtouge, marking the emergence of the "Tongyun Ge" AI triangle consisting of Tongyi Lab, Alibaba Cloud, and Pingtouge [2][4]. - The "Zhenwu" PPU features a self-developed parallel computing architecture with 96G HBM2e memory and an inter-chip bandwidth of 700 GB/s, suitable for AI training, inference, and autonomous driving [8]. Group 2: Model Performance - The Qwen3-Max-Thinking model, released by Tongyi Lab, has set multiple global records, with over 1 trillion parameters and 36 trillion tokens of pre-training data, rivaling models like GPT-5.2 and Gemini 3 Pro [10]. - The model has demonstrated superior performance in 19 benchmark tests, showcasing its capabilities in factual knowledge, complex reasoning, and alignment with human preferences [10]. Group 3: Market Impact and Adoption - The "Zhenwu" PPU has been deployed in over 400 clients, including major organizations like State Grid and Xpeng Motors, indicating strong market demand and positive reception [4][8]. - The Qwen3-Max-Thinking model is available for free on QwenChat, with over 200,000 derivative models and 1 billion downloads, establishing it as a leader in the global AI open-source community [11].

据介绍，"真武"PPU采用自研并行计算架构和片间互联技术，配合全栈自研软件栈。其内存为96G HBM2e，片间互联带宽达到700 GB/s，可应用于AI训练、AI推理和自动驾驶。 阿里巴巴已将"真武"PPU大规模用于千问大模型的训练和推理，并结合阿里云完整的AI软件栈进行深度 优化，为客户提供一体化产品和服务。 智通财经APP获悉，阿里巴巴(09988)平头哥官网上线高端AI芯片"真武810E"产品，该芯片实现软硬件 全自研，已在阿里云实现多个万卡集群部署。每颗真武810E芯片配备7个ICN片间互联端口，配合平头 哥自研互联加速库，实现多卡协同工作，从而高效支持大模型训练及推理需求。 公开资料显示，平头哥半导体有限公司成立于2018年9月，作为阿里巴巴集团的全资半导体芯片业务主 体，其拥有端云一体全栈产品系列(涵盖数据中心芯片、IoT芯片等)，实现芯片端到端设计链路全覆 盖。平头哥已推出多款数据中心及物联网芯片，包括AI芯片"PPU"、推理芯片"含光800"、CPU"倚天 710"、SSD主控芯片"镇岳510"及IoT芯片"羽阵"。 此前媒体传出阿里巴巴(09988)正准备推进旗下AI芯片制造企业"平头哥"( ...

此前媒体传出阿里巴巴(09988)正准备推进旗下AI芯片制造企业"平头哥"(T-Head)独立上市。阿里计划将 其重组为一个部分由员工持股的业务实体，随后再寻求进行首次公开募股(IPO)。 公开资料显示，平头哥半导体有限公司成立于2018年9月，作为阿里巴巴集团的全资半导体芯片业务主 体，其拥有端云一体全栈产品系列(涵盖数据中心芯片、IoT芯片等)，实现芯片端到端设计链路全覆 盖。平头哥已推出多款数据中心及物联网芯片，包括AI芯片"PPU"、推理芯片"含光800"、CPU"倚天 710"、SSD主控芯片"镇岳510"及IoT芯片"羽阵"。 据介绍，"真武"PPU采用自研并行计算架构和片间互联技术，配合全栈自研软件栈。其内存为96G HBM2e，片间互联带宽达到700GB/s，可应用于AI训练、AI推理和自动驾驶。 阿里巴巴已将"真武"PPU大规模用于千问大模型的训练和推理，并结合阿里云完整的AI软件栈进行深度 优化，为客户提供一体化产品和服务。 阿里巴巴(09988)平头哥官网上线高端AI芯片"真武810E"产品，该芯片实现软硬件全自研，已在阿里云 实现多个万卡集群部署。每颗真武810E芯片配备7个ICN片间 ...

1月29日上午，阿里巴巴（NYSE:BABA，09988.HK）旗下平头哥半导体（T-Head）官网公布高端AI芯 片"真武810E"产品信息。 官网介绍显示，该芯片搭载自研并行计算架构和ICN（Inter-Chip-Network）片间互联技术，配合全栈自 研软件栈，实现软硬件结合。 平头哥自研的ICN片间互联技术具备高性能、高带宽、低延迟等优势，适用于大模型训练和推理应用。 每颗真武810E芯片配备7个ICN片间互联端口，配合平头哥自研互联加速库，实现多卡协同工作，从而 高效支持大模型训练及推理需求。其主要应用场景包括自动驾驶、AI训练、AI推理及多模态模型。 1月22日，据多家媒体报道，阿里巴巴正计划重组旗下芯片公司平头哥半导体，并支持其未来独立上 市。截至目前，公司尚未对此事作出正式回应。 财经频道更多独家策划、专家专栏，免费查阅>> 责任编辑：栎树 目前，真武810E已在阿里云落地多个万卡集群，并服务了国家电网、小鹏汽车等400多家客户。其正式 上线，意味着由通义实验室、阿里云和平头哥组成的阿里AI黄金三角"通云哥"完整浮出水面。阿里成为 继谷歌之后全球第二家具备"大模型+云+芯片"全栈自研实力的科 ...

Core Viewpoint - Alibaba's semiconductor subsidiary, Pingtouge, has officially launched the high-end AI training and inference chip, Zhenwu 810E, marking a significant step in its AI chip strategy [1][3]. Group 1: Product Launch and Applications - The Zhenwu 810E chip has been deployed in multiple large-scale clusters on Alibaba Cloud, serving over 400 government and enterprise clients across various sectors, including energy, research, automotive, and internet [3]. - The chip is part of Alibaba's AI ecosystem, "Tongyun Ge," which integrates chip development, cloud services, and model algorithms, creating a comprehensive AI solution [3][4]. Group 2: Technical Specifications - Zhenwu 810E features a self-developed parallel computing architecture and ICN interconnect technology, equipped with 96GB HBM2e memory and a high interconnect bandwidth of 700GB/s [4]. - The chip's performance surpasses Nvidia's A800 and is on par with Nvidia's H20, with reports indicating that its upgraded version outperforms Nvidia's A100 [4]. Group 3: Core Advantages - The chip boasts six core advantages: fully self-developed architecture, high-performance computing power, strong interconnect capabilities, high usability, large-scale validation, and synergy between chip and cloud [5]. - Pingtouge's self-developed ICN technology is crucial for supporting large model training and inference, enabling efficient multi-card collaboration [5]. Group 4: Application Scenarios - Zhenwu 810E supports various application scenarios, including over 50 common models in autonomous driving, high-performance training capabilities, and optimization for large model inference [6]. - The chip has shown outstanding performance in multi-modal applications, providing cost-effective computing support for diverse scenarios [6]. Group 5: Strategic Development - Since the establishment of Alibaba Cloud in 2009 and Pingtouge in 2018, Alibaba has invested 17 years in strategic integration to complete the "Tongyun Ge" AI ecosystem [7]. - Pingtouge aims to innovate chip infrastructure for the AI era, focusing on data center chips and IoT chip development to support the digital and intelligent transformation [7].