Chinese tech firm Alibaba has announced the launch of its XuanTie C950 server chip, a new 5-nanometer processor built to handle heavier cloud and AI agent workloads. Per rumors, this processor runs at 3.2 GHz and uses the open source RISC-V architecture, and Alibaba says it is “the highest performing RISC-V CPU in the world.”
According to Alibaba, the new chip is also more than three times faster than the older XuanTie C920. In the announcement blog post, Alibaba said RISC-V’s open standard lets chip designers customize instruction sets and speed up specific AI jobs with little or no licensing fees. “RISC-V’s open-standard nature allows chip designers to customize instruction sets and accelerate specific AI workloads with no or low licensing fees. This is particularly important for the development of AI agents,” said Alibaba.
Alibaba to push agentic AI tools with new chip
The new XuanTie C950 is part of a chip plan being pushed through T-Head, Alibaba’s chip arm, which is mainly focused on the Zhenwu 810E series for AI training and inference, while the XuanTie line is aimed at high-performance cloud systems and agentic AI. The launch came days after Alibaba rolled out Wukong, an enterprise platform built for AI agent workflows, as more companies and institutions across China adopt OpenClaw.
On Monday, the firm launched Accio Work, the international version of that platform. Alibaba said the system can autonomously handle all types of business operations for small and medium-sized enterprises. Earlier this month, the company also regrouped some of its AI teams under a new unit called Alibaba Token Hub, which is focused on building AI work platforms for enterprise users.
That business change is happening as Alibaba looks for new ways to stay profitable while competition in China pushes down token prices for AI models. The pressure is simple. Cheaper model usage means companies need other ways to make money. So Alibaba is going deeper into chips, cloud systems, and enterprise software at the same time.
Meanwhile, Huawei also stepped up its own AI hardware push with the launch of the Atlas 350 accelerator card for inference, unveiled at its China Partner Conference on Friday. Huawei Vice President Ma Haixu said the card runs on the new Ascend 950PR chip, which was designed to provide stronger computing power and storage for AI inferencing.
Zhang Dixuan, head of Huawei’s Ascend computing business, said the Atlas 350 delivers 1.56 petaflops of FP4 computing power, which is 2.8 times the level of Nvidia’s China specific H20 chip. He said the product is meant to match or beat other players in search recommendation, multimodal generation, and large language models. An accelerator card is a hardware unit made for a specific task and built to be added to a server.
The Huawei launch came as the US-sanctioned company keeps expanding its AI computing infrastructure with chips it developed on its own, including its Ascend AI line, without using American technology. At the same time, pressure is building on Nvidia from Washington. In a letter to Commerce Secretary Howard Lutnick, Republican Senator Jim Banks and Senator Elizabeth Warren called for “immediate action” over what they called a “large-scale diversion of advanced American AI chips to China” through Southeast Asia.
