Alibaba wants to expand data center capacity past 20 GW by 2032, says its new V900 AI chip can scale to 500,000-card clusters, but will 'China's most powerful AI chip' be enough to displace Nvidia GPUs?

Alibaba, the Chinese e-commerce giant, is planning to expand its data center capacity to over 20 gigawatts (GW) by 2032, as revealed by its newly unveiled Zhenwu V900 AI chip. The company claims that the V900 can scale up to support clusters of up to 500,000 GPUs. However, the question remains whether Alibaba's chip will be enough to challenge Nvidia's Dominance in the AI GPU market.

The Zhenwu V900 features 216GB of memory and a connection speed of 1.2TB/s. Alibaba has not disclosed specific performance metrics such as FLOPS, node, foundry, or power consumption figures. Its performance is expected to fall short of Huawei's Ascend 960DT, which was released in the same quarter and has provided more detailed specifications.

Alibaba's CEO, Eddie Wu, has set a target of more than 20GW of capacity for Alibaba Cloud-operated data centers by 2032. However, this goal faces challenges due to supply chain limitations. The company has shipped over 560,000 Zhenwu chips to date, but maxing out a single V900 cluster would require nearly the same number of chips.

The V900's high GPU count and memory capacity come with significant power requirements. If each V900 GPU consumes approximately 700W, as reported by Nvidia for their H200 GPUs, the cluster could draw a substantial amount of power. With China's reliance on older process nodes due to US sanctions, the power consumption of a 500,000-GPU cluster would be considerable.

Alibaba's CEO, Eddie Wu, acknowledged that demand for AI infrastructure continues to outpace supply, leading to potential supply chain limitations. The company aims to achieve a fivefold increase in AI data center capacity over the next six years, with Alibaba Cloud's current footprint estimated at around 4GW.