Alibaba outlined a new AI hardware and model roadmap at its Apsara conference in Hangzhou on September 22, 2026, according to Reuters. The announcements cover chips, models and the infrastructure required to run them.
A chip announcement with production still ahead
The company introduced the Zhenwu V900, developed by its T-Head unit, with mass production expected in early 2027. Alibaba says it delivers three times the performance of its predecessor. That comparison is a company claim, not an independent benchmark result.
Model and infrastructure ambitions
Alibaba also plans a model with five to ten trillion parameters. Associated Press reports that its current Qwen3.8-Max has 2.4 trillion parameters and that the company is targeting data-centre capacity of 20 gigawatts by 2032.
The announcements describe a development roadmap: the future chip, model and capacity targets should not be read as resources already available to customers. Larger parameter counts alone also do not demonstrate better performance on a particular task.
The next substantive milestones will be production availability, published evaluation results and delivered infrastructure. Those will provide firmer evidence for assessing how the plans translate into usable AI capacity.
Featured image: AI-generated editorial illustration; not a documentary photograph.
