Alibaba launched the Zhenwu V900 artificial intelligence accelerator, which is positioned as “the most powerful artificial intelligence chip in China today.” According to the company, the new accelerator has three times the performance of its predecessor, the Zhenwu M890.
Image source: ithom
The Zhenwu V900 AI accelerator was developed by T-Head Semiconductor. It is equipped with 216GB HBM memory, supports FP8/FP4 computing, and provides 1.2TB/s inter-chip interconnect bandwidth. According to Gao Hui, vice president of the flat head display, the product is scheduled to begin mass production and be launched in the first quarter of 2027. The V900-based computing cluster can accommodate up to 500,000 chips and is used to solve training and inference problems.
T-Head also announced that it is preparing the next-generation AI accelerator Zhenwu J900, which is scheduled to be released in the third quarter of 2028. The department also plans to build its own new generation of central processing units (CPUs) optimized for agent-based artificial intelligence tasks – the Etian 720 and Etian 730, expected to be released in 2027. It is reported that Yitian 730 will be the first T-Head processor based on self-developed microarchitecture.
Alibaba CEO Wu Yongming also introduced the company’s roadmap, emphasizing that “AI models, AI chips, and AI cloud” are the three cornerstones of the era of machine intelligence.
As part of artificial intelligence (ASI) development, the Qwen development team plans to train Qwen 5 models with 5-10 trillion parameters, designed to perform complex tasks for long periods of time.
By comparison, Alibaba’s flagship Qwen 3.8-Max has 2.4 trillion parameters, while Moonshot AI’s Kimi K3 (China’s largest open model) has 2.8 trillion parameters.
In addition, Alibaba has launched Qwen Intelligence, a comprehensive platform for creating AI agents that will provide smartphone manufacturers with tools powered by Qwen.
If you find an error, select it with your mouse and press CTRL+ENTER.