Alibaba Unveils Zhenwu V900 AI Chip, Targets 10-Trillion-Parameter Qwen

At Apsara 2026, Alibaba unveiled the Zhenwu V900 AI chip and outlined Qwen models scaling to 5-10 trillion parameters, with a 20-GW data-center target by 2032.

Alibaba Unveils Zhenwu V900 AI Chip, Targets 10-Trillion-Parameter Qwen

Alibaba on Monday, September 22, 2026, used the opening keynote of its annual Apsara Conference in Hangzhou to unveil Zhenwu V900, an in-house AI accelerator that the company positioned as China's most powerful chip yet, and to lay out plans for Qwen models scaling from 5 to 10 trillion parameters.

Zhenwu V900: Triple The Performance, 500,000-Unit Clusters

The Zhenwu V900 delivers more than three times the performance of its predecessor and supports clusters of up to 500,000 units for training frontier-scale models, according to Alibaba executives at the event. Commercial availability is scheduled for early 2027, giving Chinese hyperscalers a domestic alternative as US export controls continue to tighten around Nvidia and AMD accelerators.

CEO Eddie Wu: 20 GW By 2032, Qwen To 10T Parameters

Group CEO Eddie Wu announced an ambitious data-center target of 20 gigawatts of global capacity by 2032, framing infrastructure buildout as the gating factor for the next generation of foundation models. He also set a Qwen roadmap that stretches from the current Qwen 3.8 family up to models spanning 5 to 10 trillion parameters, well beyond the 2.4T-parameter Qwen 3.8-Max released in August 2026.

Alibaba campus in Hangzhou, home of the Apsara Conference

$53 Billion, $10.2 Billion Raise, And Citi's $160 Billion Revenue Case

The chip and model roadmap sit on top of a three-year $53 billion AI investment that Alibaba first outlined earlier this year, topped up by roughly $10.2 billion raised through an additional share offering in August 2026. Citigroup analysts have modeled that Alibaba's data-center expansion alone could generate more than $160 billion in incremental revenue over the buildout window.

The Bigger Picture: A Domestic Nvidia Alternative

Zhenwu V900 lands the same week that Apple leaned on Nvidia NVLink Fusion for its M8-based AI servers and days after Microsoft open-sourced TauGrid for GPU scheduling. Alibaba's message from Hangzhou is unmistakable: a full-stack China AI platform, from silicon to trillion-parameter models, built on domestic supply.

Reporting based on coverage from Quartz, Seoul Economic Daily and The Washington Times.

Category: AI & Technology

Tags: AI Alibaba China AI Infrastructure AI Chips

Related Articles