WHY THIS MATTERS
CONFIRMED: at the September 22 Apsara Conference, Alibaba announced the Zhenwu V900 AI chip, with triple the performance of its predecessor and support for clusters of up to 500,000 units, plus a 20 gigawatt data center target by 2032 and a 5 to 10 trillion parameter Qwen4 already in training.In this article
CONFIRMED: Alibaba bets on its own chip to power the next generation of Qwen
At the Apsara Conference on September 22, Alibaba unveiled the Zhenwu V900, described by the company as the most powerful AI chip ever made in China. The announcement came bundled with an infrastructure roadmap covering chips, data centers and the next generation of Qwen models, and Alibaba shares rose the day of the presentation.
What the new chip promises
According to Alibaba, the Zhenwu V900 delivers triple the performance of the Zhenwu M890, its previous chip, with enough memory and bandwidth to support clusters of up to 500,000 units working together for training and inference. Mass production and commercial launch are planned for the first quarter of 2027, meaning today's announcement is a confirmed roadmap, not a product available right now.
A much bigger Qwen in the oven
The company confirmed Qwen4 is already in training, with Qwen4.5 and Qwen5 also in the pipeline. The stated goal is to train a next generation model in the 5 to 10 trillion parameter range, a measure of learning capacity well above anything the Qwen family has publicly shipped so far.
Energy as the declared bottleneck
Alibaba also set a target of surpassing 20 gigawatts of cloud data center capacity by 2032, citing exponentially growing demand for AI computing power. Placing such a large energy figure next to the chip announcement makes clear that, for the company, the real bottleneck to scaling AI is not just silicon, it is electrical capacity.
Why this matters outside China
A proprietary chip at scale reduces Alibaba's dependence on foreign semiconductor suppliers at a time of export restrictions between the United States and China. For the global market, Brazil included, the indirect effect worth watching is whether cheaper infrastructure for Alibaba translates into even more aggressive API pricing for the Qwen family, which already competes on cost with Western models.
MaxAssistant's read
None of this changes what a Brazilian company uses tomorrow morning. The value is in understanding the direction: Alibaba is vertically integrating chip, data center and model at the same time, betting that controlling the whole stack is a long term competitive advantage. Worth watching whether the V900's performance numbers hold up under independent testing once the chip actually reaches the market in 2027.
Sources
CNBC, Alibaba shares jump as new AI chip, data center buildout plans unveiled: https://www.cnbc.com/2026/09/22/alibaba-ai-alibabacloud-zhenwu-v900-.html | Tom's Hardware, Alibaba unveils Zhenwu V900 AI accelerator, claims it's the most powerful AI chip in China: https://www.tomshardware.com/tech-industry/artificial-intelligence/alibaba-unveils-zhenwu-v900-ai-accelerator-claims-its-the-most-powerful-ai-chip-in-china-accelerator-supports-500-000-chip-supercluster-with-a-10t-parameter-qwen-model-on-the-roadmap | Seeking Alpha, Alibaba steps up AI push with V900 chip, 20 GW data center target: https://seekingalpha.com/news/4645223-alibaba-steps-up-ai-push-with-v900-chip-20-gw-data-center-target