Alibaba's T-Head division has unveiled the Zhenwu V900, an AI accelerator designed for training and inference of large language models. The new chip offers three times the performance of its predecessor, the M890, which was released in May. The V900 features 216 GB of memory, a 1200 GB/s interconnect bandwidth, and support for FP8 and FP4 computing formats.
The company plans to begin mass production and commercial deployment of the V900 in the first quarter of 2027. Alibaba also introduced a server supernode based on the V900, allowing for the creation of clusters containing up to 500,000 chips. These solutions are currently utilized by over 650 external clients across more than 20 industries via Alibaba Cloud.
Alibaba is scaling its Qwen model family, with plans for Qwen 4.5 and Qwen 5 to reach 5–10 trillion parameters, compared to the 2.4 trillion in the current Qwen3.8-Max. The company is also integrating self-learning capabilities into its models. In a recent test, an AI model spent over 60 hours and performed more than 10,000 EDA tool calls to optimize a chip block, reducing its area by 42%.
To support its growing AI infrastructure, Alibaba aims to expand its data center capacity to over 20 GW by 2032. The company is prioritizing the development of its own hardware amid US export restrictions on advanced Nvidia chips. In August, Alibaba raised $10.2 billion to invest in computing infrastructure, proprietary chips, and AI applications.
