Alibaba's DAMO Academy Launches Xuantie C950 CPU, First to Natively Support Billion-Parameter AI Models

RISC-V architecture is officially transitioning from a "robust alternative" to a "leading force" in the computing market.

On March 24, at the 2026 Xuantie RISC-V Ecosystem Conference in Shanghai, Alibaba's DAMO Academy officially unveiled the high-performance RISC-V CPU, the Xuantie C950. This processor's launch not only set a new global performance benchmark for RISC-V but also marked the first CPU with native support for trillion-parameter large language models.

Performance Breakthrough: Single-Core Score Surpasses 70

As the latest flagship from DAMO Academy, the Xuantie C950 features a significantly upgraded core architecture:

Core Specifications: It employs an 8-wide instruction decode, a 16-stage pipeline, and a reorder buffer supporting over 1000 instructions, achieving a maximum frequency of 3.2GHz.

Performance Record: In SPECint2006 benchmark testing, its single-core general-purpose performance score exceeded 70 for the first time, establishing it as the world's most powerful RISC-V processor.

Efficiency and Performance: In typical workloads like cloud networking and storage, its performance surpasses mainstream products by over 30%.

Native AI Acceleration: A New Home for Trillion-Parameter Models

Beyond exceptional general-purpose computing power, DAMO Academy also introduced two native AI acceleration engines: Vector and Matrix.

These engines share a unified address space with the Xuantie CPU, enabling deep integration of general and AI computing capabilities. Testing confirms this architecture can smoothly run leading large models like Qwen3-235B and DeepSeek V3-671B. This industry-first native CPU support for trillion-parameter models provides a new hardware blueprint for servers in the AI Agent era.

Ecosystem Expansion: Targeting 36 Billion Units Shipped

At the conference, Chinese Academy of Engineering Academician Ni Guangnan stated RISC-V is rapidly approaching mainstream adoption. Industry data projects RISC-V device shipments will reach 3.6 billion by 2031.

To accelerate this growth, DAMO Academy is actively engaged:

Strategic Partnerships: Signed agreements with the Chinese Academy of Sciences' Institute of Software and the Beijing Open Source Chip Research Institute to co-develop the next-generation open-source Xiangshan CPU.

System Compatibility: Collaborating with openKylin to build a RISC-V-native AI operating system, with support for OpenClaw deployment.

Mass Production Momentum: The Xuantie series is already integrated into over 200 mass-produced chips, widely used in cutting-edge fields like AI glasses and industrial power control.

From edge devices to high-performance servers, the technological prowess represented by Alibaba