A server processor tailored for AI, handles massive models and dynamic workloads enabling faster processing and better resource utilization across data center environments
Alibaba Group has unveiled its latest server-class CPU, the XuanTie C950, targeting AI agents and high-performance cloud workloads as demand for large-scale AI models accelerates. Developed under its T-Head semiconductor division, the processor is based on the open-source RISC-V architecture and has reportedly been verified on a 5 nm process.
The launch comes amid a surge in AI agent deployment across China, driven by increasingly complex models such as Qwen3 and DeepSeek V3. As workloads scale to hundreds of billions of parameters, there is growing need for processors that combine flexibility, performance and ecosystem independence beyond traditional x86 and Arm architectures.
The CPU addresses these demands by integrating a proprietary AI acceleration engine and enabling native support for large-scale inference workloads. This positions RISC-V as a viable contender in high-end AI computing, an area traditionally dominated by established instruction sets.
From a performance standpoint, the chip features an eight-instruction decode, a 16-stage pipeline, and an out-of-order execution window exceeding 1,000 instructions. It achieves a SPECint2006 score above 70 per core and operates at clock speeds up to 3.2 GHz, enabling it to handle mainstream cloud and server tasks. Alibaba also claims over 30 percent performance gains in certain workloads, driven by its ability to tailor the CPU architecture to specific inference patterns.
For system designers and cloud providers, the CPU offers advantages in customization and supply chain resilience. Its open architecture allows workload-specific optimization, while reducing dependence on constrained global compute ecosystems.
The CPU also marks a strategic step toward positioning RISC-V at the core of next-generation AI infrastructure.
Click here for the official announcement.
