Alibaba unveils Zhenwu V900, sets 20 GW-plus cloud capacity target

Alibaba unveiled its Zhenwu V900 AI accelerator at the Apsara Conference in Hangzhou on September 22, setting a target for the global data-center capacity operated by Alibaba Cloud to exceed 20 gigawatts by 2032.

The 20 GW figure is a company target, not installed or independently verified capacity. The opened sources do not specify whether it refers to total facility power, IT equipment load, contracted utility capacity or another measure. They also provide no comparable current baseline or annual construction milestones.

Chief Executive Eddie Wu called the V900 China’s most powerful AI chip and said it delivers three times the performance of the preceding Zhenwu M890. None of the opened sources provides independent benchmark results or test methodology to support either claim. The sources also do not identify the chip’s fabrication process, foundry, power draw, production volume, price or delivery schedule.

Alibaba said a V900-based cluster could scale to as many as 500,000 chips for training and running frontier AI models. That is an announced ceiling: the opened sources do not document a deployed cluster of that size or provide networking and reliability results.

Alibaba presented models, chips and cloud capacity as three linked layers of its AI infrastructure investment plan. Wu said strong customer demand for AI computing and shortages across the global AI data-center supply chain were limiting how quickly Alibaba Cloud could expand. The company also said it plans to train a model with 5 trillion to 10 trillion parameters, but did not say training had begun.

The official conference program scheduled Wu’s keynote and a presentation by T-Head Vice President Gao Hui. The opened sources do not provide site locations, financing details or a capital-expenditure allocation for the data-center target.