Alibaba used its Apsara Conference to announce a full-stack AI roadmap spanning models, chips, cloud infrastructure and agent products. The company said Qwen 4 is in training, outlined future Qwen 4.5 and Qwen 5 systems, and introduced T-Head’s Zhenwu V900 accelerator.
A new accelerator and supernode roadmap
Alibaba says the Zhenwu V900 has 216 GB of memory, 1,200 GB/s of inter-chip bandwidth and three times the performance of the earlier M890. It says an integrated supernode architecture could scale to clusters of as many as 500,000 cards, with mass production and commercial release planned for the first quarter of 2027.
Qwen plans and a 20-gigawatt cloud target
The company says Qwen 4 is currently in training and that future Qwen 4.5 and Qwen 5 systems are projected to reach 5 trillion to 10 trillion parameters. Alibaba also set a target for the data-center capacity operated by Alibaba Cloud to exceed 20 gigawatts by 2032.
Reuters independently reported the conference announcements, but the chip’s performance, the proposed model sizes and the 2032 infrastructure target have not been independently demonstrated. The V900 is not yet in commercial mass production, and Alibaba did not disclose all specifications needed for direct comparisons with rival accelerators.