Alibaba teases 10-trillion-parameter model, debuts ‘China’s most powerful’ AI chip

Alibaba Group Holding unveiled a new AI chip and announced ambitious plans to train an AI model with up to 10 trillion parameters, positioning itself as "China's most powerful" AI entity. At its annual Apsara Conference in Hangzhou, group chairman Joe Tsai emphasized Alibaba's commitment to building end-to-end AI capabilities, aiming to move from technological breakthroughs to value creation.

CEO Eddie Wu Yongming outlined three key areas: AI models, AI chips, and AI cloud. The company targets scaling its global data-center capacity to over 20 gigawatts by 2032. Alibaba's Qwen development team plans to train Qwen 5, with 5 trillion to 10 trillion parameters, to tackle complex tasks over extended periods. The current flagship model, Qwen 3.8-Max, boasts 2.4 trillion parameters, while Moonshot AI's Kimi K3 leads with 2.8 trillion parameters.

Liu Dayiheng, head of the Qwen LLM project, presented the roadmap for Qwen 4, Qwen 4.5, and Qwen 5, focusing on advancements in recursive self-improvement and enhancements across speech, vision, and multimodal models. Alibaba also launched Qwen Intelligence, a full-stack agent platform for smartphone makers to access Qwen-powered tools for AI-enabled phones.

The company confirmed collaboration with Apple on AI features for iPhones in China. Alibaba debuted the Zhenwu V900 processor, claiming it as the "most powerful AI chip in China today," delivering three times the performance of its predecessor. The chip can link up to 500,000 chips for training and inference of frontier models. T-Head, Alibaba's semiconductor subsidiary, unveiled the Zhenwu V900 and previewed the next-generation Zhenwu J900 and CPUs tailored for agentic AI tasks, with commercial availability slated for the first quarter of 2027.

Despite US export controls, China continues to train its most advanced AI models on Nvidia processors. Alibaba's Hong Kong-listed shares rose 3% after the announcements.