阿里巴巴集团控股有限公司的研究院达摩院表示，基于其玄铁RISC-V架构的芯片出货量已超过60亿颗，标志着该科技巨头在人工智能和汽车系统领域加大对开放标准的投入取得重要里程碑。
Alibaba Group Holding’s research arm, Damo Academy, has seen more than 6 billion chips shipped based on its XuanTie RISC-V architecture, marking a milestone as the tech giant doubles down on the open standard in artificial intelligence and automotive systems.
达摩院首席科学家孟建毅周一在深圳举办的年度RISC-V中国峰会上公布了这一数据。此次成果的取得正值中国企业力求在全球RISC-V生态系统中发挥更大作用之际，该生态系统是英特尔和Arm等公司主导的传统芯片架构的替代方案。
Meng Jianyi, chief scientist at Damo Academy, announced the figure on Monday at the annual RISC-V Summit China in Shenzhen. The achievement comes as Chinese companies seek to play a greater role in the global RISC-V ecosystem, an alternative to established chip architectures developed by the likes of Intel and Arm.
总部位于杭州的阿里巴巴是中国RISC-V领域的早期参与者之一，目前正在扩展其人工智能基础设施能力。由于该架构是开放标准，开发者可以利用它设计处理器，从而在美国主导的出口限制背景下，有望减少对外国芯片设计技术的依赖。
Alibaba, based in Hangzhou, was an early mover in China’s RISC-V landscape and is expanding its AI infrastructure capabilities. Because the architecture is an open standard, developers can use it to design processors while potentially reducing reliance on foreign chip-design technologies amid US-led export restrictions.
为凸显该架构的快速全球普及，孟建毅引用了研究机构Omdia的预测，称全球RISC-V处理器出货量预计年增近30%，到2030年有望占全球处理器市场的约25%。
Highlighting the architecture’s rapid global adoption, Meng cited forecasts from research firm Omdia that global RISC-V processor shipments would grow by nearly 30 per cent annually and could account for about 25 per cent of the worldwide processor market by 2030.
这一出货里程碑紧随今年早些时候玄铁C950的发布——阿里巴巴曾称其为全球最强大的同类RISC-V处理器。公司表示，该芯片的开发旨在支持云计算和高性能计算环境中的AI代理及其他工作负载，其性能较前代产品C920提升了三倍以上。
The shipment milestone follows the launch earlier this year of the XuanTie C950, a flagship RISC-V processor that Alibaba called the most powerful of its kind globally. The company said it was developed to support AI agents and other workloads in cloud and high-performance computing environments, and that it delivered more than three times the performance of its predecessor, the C920.
C950是中国科技公司推动构建AI所需计算基础设施的更广泛举措的一部分。阿里巴巴、腾讯控股和字节跳动也在竞相开发AI代理——即能够代表用户或其他系统自主执行任务的软件——并将其集成到各自的应用程序中。
The C950 is part of a broader push by Chinese tech companies to develop the computing infrastructure needed for AI. Alibaba, Tencent Holdings and ByteDance are also competing to develop AI agents – software that can autonomously perform tasks on behalf of users or other systems – and integrate them into their apps.
RISC-V，即第五代精简指令集计算机，是一种开放标准的指令集架构，允许开发者设计中央处理单元（CPU）。它由非营利组织RISC-V International管理。
RISC-V, short for fifth-generation reduced instruction set computer, is an open-standard instruction set architecture that allows developers to design central processing units (CPUs). It is managed by the non-profit RISC-V International.
RISC-V对任何人免费使用和修改，被视为中国实现芯片设计更高自给自足的潜在途径。该架构为设计者提供了Intel x86和Arm同名架构的替代方案，前者为运行Microsoft Windows的设备提供动力，后者为运行Google Android的设备提供动力。
Free for anyone to use and modify, RISC-V is considered a potential route for China to achieve greater self-sufficiency in chip design. The architecture offers designers an alternative to Intel’s x86 and Arm’s namesake architecture, which power devices running Microsoft Windows and Google Android, respectively.
然而，孟某指出，尽管取得进展，但阿里巴巴在提升其RISC-V CPU在运行大型语言模型时的性能过程中仍面临工程挑战。
Despite the progress, however, Meng highlighted the engineering challenges involved as Alibaba seeks to improve the performance of its RISC-V CPUs when running large language models.
“最大的挑战在于协同设计，”孟某在接受RISC-V International采访时表示，该采访由达摩学院上周发布。“孤立的原始算力意义不大；关键在于平衡计算、数据移动和软件优化。”他补充说，对于Qwen3.8-27B等大型AI模型，量化、矩阵指令和内存带宽等技术必须作为一个统一系统进行集成。阿里巴巴拥有《南华早报》。
“The biggest challenge lies in co-design,” Meng said in an interview with RISC-V International, released by the Damo Academy last week. “Isolated raw computing power means little; the key is balancing computation, data movement and software optimisation.”For large AI models such as Qwen3.8-27B, he added, techniques such as quantisation, matrix instructions and memory bandwidth must be integrated as a unified system. Alibaba owns the South China Morning Post.