Alibaba Aims for 500,000 Connected Chips and 20 GW by 2032

Key Takeaways

1Alibaba unveils the Zhenwu V900 chip, announced to be three times more powerful than the M890

2The chip can be connected in clusters of up to 500,000 processors

3Goal of 20 GW data center capacity by 2032 and scaling up of Qwen models

4A Trump–Xi meeting and the presence of Jensen Huang are anticipated

Why it matters : This announcement highlights Alibaba's ambition to compete with American cloud and AI giants despite U.S. restrictions.

The Chinese group unveils the Zhenwu V900, its AI chip touted to be three times more powerful than the M890 and intended for mass production in the first quarter of 2027. Meanwhile, Alibaba is pushing for very large cluster architectures, planning 20 GW of data center power by 2032, and accelerating the Qwen model family.

Clusters of up to 500,000 processors and 20 GW targeted

The Zhenwu V900 has been designed to be connected in very large sets, with up to 500,000 interconnected processors. Alibaba is also setting an infrastructure target: 20 gigawatts of cumulative capacity for its data centers by 2032. These choices align with international ambitions in the cloud, competing with AWS, Microsoft Azure, and Google Cloud.

The Zhenwu V900, technical specifications and production timeline

Eddie Wu, CEO of Alibaba, presented the Zhenwu V900 in Hangzhou during the Apsara conference. The company describes it as the most powerful Chinese AI chip to date and announces a tripled power compared to the M890 model. Large-scale commercialization is expected in the first quarter of 2027.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Alibaba connects numerous processors to bypass U.S. restrictions

Due to restrictions imposed by the United States, which limit access to advanced technologies, Alibaba opts for the combination of multiple processors instead of relying solely on the increased performance of a single chip, contrary to the strategy typically associated with NVIDIA. Huawei is taking a similar approach, and its Ascend AI chips are gradually gaining ground in the Chinese market.

Qwen roadmap and imminent diplomatic context

On the software side, Alibaba is already training the Qwen 4 generation and preparing the Qwen 4.5 and Qwen 5 series. The upcoming iterations aim for a scale between 5 and 10 trillion parameters, while Qwen 3.8 Max claims 2.4 trillion. This sequence comes on the eve of a scheduled meeting between Donald Trump and Xi Jinping, where NVIDIA's leader, Jensen Huang, is also expected to attend.