Alibaba unveiled its latest artificial intelligence chip, the Zhenwu V900, as the Chinese technology major expands its push to build a full-stack AI ecosystem. Developed by Alibaba’s semiconductor unit T-Head, the new accelerator is designed for both AI training and inference.

Alibaba claims the V900 delivers three times the performance of its previous-generation Zhenwu M890 chip, which was launched in May. The company did not provide independent benchmark results for the performance claim. The V900 is scheduled to enter mass production and commercial release in the first quarter of 2027.

The V900 comes with 216GB of memory and 1,200GB/s of chip-to-chip bandwidth. It also supports FP8 and FP4 precision formats, which are increasingly important for efficient AI inference and large-scale model workloads.

Alibaba says, “The chip can be connected into clusters containing as many as 500,000 accelerators. This architecture is designed to handle demanding AI workloads that require thousands of processors to work together.”

The company is also integrating the V900 with its broader hardware stack, including networking and storage components. Thus, allowing Alibaba to control more parts of the infrastructure required to train and deploy large AI models. The strategy links Alibaba’s chip development directly with its AI model roadmap. More powerful accelerators can provide the computing capacity needed to train increasingly large models.

Alibaba is also planning a major expansion of its cloud infrastructure. CEO Eddie Wu said, “Alibaba Cloud aims to increase its global data-centre capacity to more than 20 gigawatts by 2032.” 

The V900 highlights Alibaba’s effort to develop chips, AI models and cloud infrastructure as one integrated ecosystem.

Source link