Alibaba Group Holding has unveiled what it claims to be China’s most powerful artificial intelligence (AI) chip and outlined plans to develop an AI model featuring up to 10 trillion parameters as part of its broader ambition to achieve artificial superintelligence (ASI).
The announcements were made during Alibaba’s annual Apsara Conference held in Hangzhou on September 19-20. Group chairman Joe Tsai emphasized the company’s focus on building a comprehensive AI ecosystem, stating the importance of transforming AI breakthroughs into practical value across industries.
Chief Executive Officer Eddie Wu Yongming outlined Alibaba’s strategy centered on three key areas: AI models, AI chips, and AI cloud infrastructure. The company aims to expand its Alibaba Cloud global data-center capacity to more than 20 gigawatts by 2032 to support the increasing computational demands of AI development.
Alibaba’s Qwen development team is targeting the training of the Qwen 5 model with 5 trillion to 10 trillion parameters. This marks a significant scale-up from the current flagship Qwen 3.8-Max, which contains 2.4 trillion parameters and ranks as the second-highest scoring Chinese model on the Artificial Analysis Intelligence Index. By comparison, Moonshot AI’s Kimi K3 holds 2.8 trillion parameters.
Liu Daiyiheng, head of the Qwen large language model (LLM) project and a core researcher since joining Alibaba in 2021, presented the road map for the Qwen series, including versions 4, 4.5, and 5. He highlighted advancements in recursive self-improvement through empirical feedback and upgrades across the Qwen models’ speech, vision, and multimodal capabilities.
In addition to model development, Alibaba introduced Qwen Intelligence, a full-stack agent platform designed to enable smartphone manufacturers to build AI-powered applications capable of managing tasks across multiple programs. The company also confirmed ongoing collaboration with Apple to incorporate AI features into iPhones sold in China.
On the hardware front, Alibaba launched the Zhenwu V900 processor, described as China’s most powerful AI chip to date. The chip is reported to deliver triple the performance of its predecessor, the Zhenwu M890. The V900-based computing cluster can link up to 500,000 chips to handle large-scale AI model training and inference. Mass production of the V900 is expected to begin in the first quarter of 2027.
T-Head, Alibaba’s semiconductor division, revealed a roadmap extending to 2028, including the introduction of the next-generation Zhenwu J900 chip in the third quarter of that year, as well as two new proprietary central processing units (CPUs) — Yitian 720 and Yitian 730 — slated for release next year. The Yitian 730 will mark T-Head’s first CPU built on a fully in-house microarchitecture.
Alibaba faces increasing competition from domestic rivals such as Huawei Technologies, which recently launched its Atlas 960 SuperPoD computing cluster powered by Ascend 960 chips. Huawei anticipates a significant shift within China toward its AI computing infrastructure for model training in the coming year, according to rotating chairman Eric Xu Zhijun.
Despite U.S. export restrictions aimed at limiting China’s access to advanced semiconductor technology, many of the country’s leading AI models reportedly continue to rely on Nvidia processors, according to previous industry sources.
Alibaba is also accelerating the expansion of its power-hungry infrastructure to meet rising computing requirements. Wu reiterated the company’s commitment to collaborating with partners to develop AI infrastructure and build an “AI cloud” tailored to the machine intelligence era.