Alibaba unveils ambitious AI roadmap with 5-10 trillion parameter models

At the 2026 Yunqi Conference in Hangzhou, Alibaba Group outlined its most ambitious AI vision, with CEO Wu Yongming announcing the development of models ranging from 5 trillion to 10 trillion parameters. These models, part of the Qwen series, aim to handle complex tasks and move toward artificial superintelligence, surpassing the current Qwen 3.8 Max with 2.4 trillion parameters. Wu highlighted recursive self-improvement, allowing models to identify their own limits and optimize themselves. Alibaba also launched the Zhenwu V900 chip, developed by TSMC, boasting three times the computing power of its predecessor, the M890. The chip is expected to mass-produce by Q1 2027. The company plans to expand its global data center capacity to over 20 gigawatts by 2032, addressing supply constraints in AI infrastructure. Wu compared the current AI shift to the Industrial Revolution, predicting machine thinking could surpass human thinking by over 1000 times. Amid U.S. export restrictions, Alibaba is advancing its AI ecosystem with super-large models, self-developed chips, and data centers, signaling a new phase in global AI competition.