Zhenwu V900: Alibaba’s most powerful chip links 500,000 units, eyes 10-trillion-parameter AI

Alibaba Group Holding has introduced what it called China’s most powerful artificial intelligence chip. The company also teased plans to train a new AI model with up to 10 trillion parameters, as top executives restated its ambition to pursue artificial superintelligence.

Speaking at the company’s annual Apsara Conference in Hangzhou, group chairman Joe Tsai stressed Alibaba’s commitment to building complete, end-to-end AI capabilities. He said the company is investing heavily in full-stack AI to help the technology move from a breakthrough into real value creation.

Three-part AI roadmap

CEO Eddie Wu Yongming laid out the company’s strategy in a keynote address. He identified AI models, AI chips and the AI cloud as the three pillars of what he called the machine intelligence era. Alibaba aims to expand its global data center capacity to more than 20 gigawatts by 2032.

Wu said machines are increasingly becoming the primary force behind thinking as AI spreads into every domain, turning intelligence into a commodity supplied at scale. To move toward artificial superintelligence, Alibaba’s Qwen team plans to train its next flagship model, Qwen 5, with 5 trillion to 10 trillion parameters, designed to handle complex tasks over extended periods.

The company’s current flagship model, Qwen 3.8-Max, holds 2.4 trillion parameters and ranks as the second-highest-scoring Chinese system on a widely cited AI benchmark. That places it just behind Moonshot AI’s Kimi K3, China’s largest open-weight model, which carries 2.8 trillion parameters.

New chip debuts

On the hardware side, Alibaba debuted its Zhenwu V900 processor, describing it as the most powerful AI chip in China today. The company said the new chip delivers three times the performance of its predecessor, the Zhenwu M890, and that a computing cluster built around it can link up to 500,000 chips for both training and inference on frontier models.

Wu said Alibaba expects a significant increase in annual AI chip shipments, though he gave no specific figure. Gao Hui, vice president of Alibaba’s semiconductor unit T-Head, said the V900 will carry 216 gigabytes of high-bandwidth memory and inter-chip bandwidth of 1.2 terabytes per second, with mass production slated for the first quarter of 2027.

Models and roadmap

Liu Dayiheng, who leads the Qwen large language model project, presented a broader roadmap covering the Qwen 4, Qwen 4.5 and Qwen 5 series in his most prominent public appearance since taking over the project. He pointed to progress in recursive self-improvement through empirical feedback, along with upgrades across Qwen’s speech, vision and multimodal capabilities.

Alibaba also launched Qwen Intelligence, a full-stack agent platform meant to give smartphone makers the tools to build AI phones capable of carrying out tasks across multiple apps.

T-Head separately outlined a hardware roadmap extending to 2028, including the next-generation Zhenwu J900 chip and two new processors, the Yitian 720 and Yitian 730, both planned for 2027. The Yitian 730 will be the company’s first CPU built on an entirely in-house microarchitecture.

The announcements come as Alibaba faces growing competition from Huawei Technologies, which recently unveiled its own Atlas 960 SuperPoD computing cluster built around its Ascend 960 chips.