Alibaba Unveils China's Most Powerful AI Chip, Plans 10 Trillion-Parameter Model
Alibaba Group Holding, a Big Tech firm, introduced what it described as China's most powerful artificial intelligence chip and outlined plans for an AI model featuring up to 10 trillion parameters. This development comes as executives reaffirmed the organisation's ambition to pursue artificial superintelligence, known as ASI. The company also aims to expand its Alibaba Cloud global data centre capacity to more than 20 gigawatts by 2032.
Group chairman Joe Tsai delivered remarks at the annual Apsara Conference in Hangzhou, underscoring Alibaba's commitment to building end to end capabilities. He stated the company is firmly investing in constructing full stack AI, intending for AI to move from technological breakthroughs to value creation.
And chief executive Eddie Wu Yongming presented the company's roadmap, identifying AI models, AI chips, and the AI cloud as the three central components of the machine intelligence era. Mr Wu also noted that as AI capabilities expand and enter every domain, a profound shift is underway where machines are becoming the main force behind thinking, thereby turning intelligence into a widely available commodity.
Alibaba's Qwen development team plans to train its Qwen 5 AI model with between 5 trillion and 10 trillion parameters. This model is designed to execute complex tasks over extended periods. The current flagship model, Qwen 3.8 Max, features 2.4 trillion parameters and holds the second highest scoring position among Chinese systems on the Artificial Analysis Intelligence Index. This compares to Moonshot AI's Kimi K3, China's largest open weight model, which has 2.8 trillion parameters.
Liu Dayiheng, head of the Qwen large language model project at Alibaba Token Hub, detailed the roadmap for the Qwen 4, Qwen 4.5, and Qwen 5 series. Mr Liu has served as a core researcher for the Qwen line since joining Alibaba some years ago, and this presentation marked his most significant public appearance in his current role.
But Liu also reported progress in recursive self improvement through empirical feedback, alongside upgrades across Qwen's speech, vision, and multimodal model families. In parallel with its frontier model development, Alibaba launched Qwen Intelligence, a full stack agent platform intended to provide smartphone makers with Qwen powered tools for developing AI phones capable of executing tasks across various applications. The organisation previously confirmed its collaboration with Apple on AI features for iPhones sold in China.
On the hardware front, Alibaba debuted the Zhenwu V900 processor, calling it China's most powerful AI chip today. The company stated this new chip provides three times the performance of its predecessor, the Zhenwu M890. A computing cluster built around the V900 can link up to 500,000 chips to manage both training and inference for frontier models, according to company statements.
Mr Wu indicated that Alibaba anticipates a significant increase in annual AI chip shipments, though he did not specify a target number. Alibaba's semiconductor subsidiary T-Head also presented its roadmap, which extends to 2028.
So Gao Hui, vice president of T-Head, stated the Zhenwu V900 would incorporate 216 gigabytes of high bandwidth memory and an inter chip bandwidth of 1.2 terabytes per second. Mass production and commercial availability for this chip are scheduled for the first quarter of 2027. T-Head also previewed the forthcoming Zhenwu J900, which is scheduled for release in the third quarter of 2028, and announced plans for next generation proprietary central processing units, the Yitian 720 and Yitian 730, both due in 2027. The Yitian 730 is expected to be T-Head's first CPU built on an in house microarchitecture.
Alibaba encounters competition from domestic rivals, including Huawei Technologies. Huawei recently unveiled its new Atlas 960 SuperPoD computing cluster, powered by its Ascend 960 chips. Huawei's rotating chairman, Eric Xu Zhijun, expressed expectations for a substantial domestic shift towards its AI computing infrastructure for model training in the coming year.
A recent report from the South China Morning Post indicated that, despite US export controls intended to limit China's access to advanced chips, China's most advanced AI models continue to be trained on Nvidia processors, citing sources at major Chinese AI developers. Alibaba stated it is also accelerating the expansion of its power intensive infrastructure to accommodate rising computing demands. Mr Wu concluded that Alibaba would join forces with partners to commit fully to AI infrastructure development, forging an AI cloud specifically designed for the machine intelligence era. Alibaba's Hong Kong listed shares closed 3 per cent higher at HKD 166 at midday.
Alibaba introduced the Zhenwu V900 processor, labelled China's most powerful AI chip.
The company plans a Qwen 5 AI model with up to 10 trillion parameters.
Alibaba Cloud aims for over 20 gigawatts of data centre capacity by 2032.
New chips and CPUs from T-Head are scheduled for release between 2027 and 2028.
The firm is collaborating with Apple on AI features for iPhones sold in China.
Source: SCMP
