Alibaba's Wu Yongming Unveils AI Roadmap: Qwen to Reach 5-10 Trillion Parameters, Zhenwu V900 Triples Performance

动察 Beating AI News Flash: Alibaba CEO Eddie Wu unveiled the next-phase AI roadmap at the Yunqi Conference:

The Qwen team plans to train new models with 5 trillion to 10 trillion parameters, T-Head has released the Zhenwu V900, and Alibaba Cloud plans to expand its global data center scale to over 20GW by 2032.

Qwen is exploring RSI (Recursive Self-Improvement), allowing models to identify their own weaknesses based on real-task feedback, design experiments, construct data, and continue training. The new-generation models will focus on enhancing complex long-horizon task capabilities while continuing to advance multimodal models that integrate understanding and generation.

Eddie Wu judges that the total amount of thinking generated by machines today is still less than 3% of that of humans, and in the future may exceed humans by 1,000 times. He believes that current AI Coding is still like "the electric light in 1882," mainly replacing existing work, and that truly new products belonging to the era of machine intelligence have not yet appeared. In the future, complex tasks may be broken down into tens of millions of subtasks and handed to millions of Agents for continuous execution.

To support computing demand on this scale, Alibaba is also simultaneously strengthening chips and cloud. The Zhenwu V900 achieves 3 times the performance of the M890, and a single cluster can scale up to 500,000 cards. Eddie Wu stated that medium- and long-term AI demand currently already exceeds supply, and Alibaba Cloud plans to operate more than 20GW of global data centers by 2032.