Alibaba Unveils Zhenwu V900 AI Chip and Plans 5–10 Trillion-Parameter Model

Alibaba CEO Eddie Wu unveiled the Zhenwu V900 AI chip, a 5–10 trillion-parameter model plan and a 20GW data center target at Apsara 2026. Shares jumped 5%.

Alibaba delivered its most aggressive roadmap yet for the artificial intelligence era on Tuesday, unveiling a next-generation AI chip, teasing a model of unprecedented scale, and pledging to build one of the world's largest computing networks. The announcements came from CEO Eddie Wu's keynote at the 2026 Apsara Conference, Alibaba Cloud's annual flagship event in Hangzhou, and investors responded immediately: shares in the Chinese e-commerce and cloud giant jumped roughly 5 percent, according to Reuters.

A 5–10 Trillion-Parameter Model in the Works For more context on this story, see our ongoing AI industry coverage.

The headline commitment is scale. Wu said Alibaba plans to train a new AI model at the scale of 5 to 10 trillion parameters, a leap well beyond the roughly 2.4-trillion-parameter Qwen3.8-Max the company has shipped to date. The model, according to Wu, is aimed at completing more complex, longer-horizon tasks and advancing toward what he called artificial superintelligence, or ASI.

A central piece of that push is Recursive Self-Improvement (RSI), a process in which models engage with real-world tasks and feedback, identify their own limitations, and autonomously design experiments, synthesize data and evaluate outcomes in a continuous cycle. Wu said Alibaba's Qwen team is already exploring RSI and has made meaningful progress, alongside continued research into model architecture and data optimization.

He also argued that multimodal models that unify understanding and generation represent a core research direction, saying that to align with human culture, aesthetics and values, models must comprehend and communicate human signals such as voice, visuals, facial expressions and gestures.

Framing the moment, Wu claimed that machine thinking currently amounts to less than 3 percent of all human thinking, while machines already drive 99.9 percent of the world's physical work. He predicted that machine thinking will ultimately shoulder 99.9 percent of all thinking, with machines producing more than 1,000 times the output of all humanity combined.

Zhenwu V900: 'The Most Powerful AI Chip in China'

On the silicon side, Alibaba's semiconductor unit T-Head introduced the Zhenwu V900, which Wu called "the most powerful AI chip in China today" — a description echoed by the South China Morning Post. The company said the V900 delivers three times the performance of its predecessor, the Zhenwu M890, and that a single cluster built on the new chip can support up to 500,000 cards for frontier model training and inference.

The V900 sits atop a broader T-Head portfolio that Wu said covers the Zhenwu series of GPU chips, the Yitian series of CPUs, the Panmai series of smart network interface cards, and ICN interconnect chips — together spanning the core silicon needed for ultra-large-scale AI clusters. The push into domestic accelerator design comes as US export controls continue to restrict Chinese firms' access to Nvidia's most advanced hardware, a dynamic that coverage from multiple outlets framed as central context for the announcement.

Alibaba also highlighted its proprietary M890 AI Supernode, which the company says already delivers high-efficiency inference for foundation models above 2 trillion parameters — a capability tier Wu claimed only a handful of companies globally hold. He said Alibaba Cloud is bringing AI Supernodes online at commercial scale and pursuing end-to-end co-optimization across chips, servers, supernodes, networks, models and inference engines.

20 Gigawatts of Data Center Capacity by 2032

The third pillar of Wu's keynote was infrastructure. He set a target for Alibaba Cloud to operate more than 20 gigawatts of global data center capacity by 2032, describing the build-out as a response to exponentially rising demand for AI compute. Bloomberg highlighted the same figure, noting the chip is designed to drive that expansion.

Wu said customer demand for AI remains exceptionally robust, and that the industry's mid-to-long-term demand far outpaces Alibaba's supply capability. He cited global shortages across the AI data center supply chain as the current limiting factor on how fast the company can scale, while saying that provisioning AI compute is driving accelerating revenue growth at Alibaba Cloud.

What It Means for the AI Race

The announcements position Alibaba as one of the very few companies attempting to build a vertically integrated AI stack — its own models, its own chips and its own cloud — alongside the likes of Google, Amazon and Microsoft. Wu described those three layers, AI models, AI chips and the AI cloud, as the infrastructure cornerstones of what he called the Machine Intelligence era.

For developers, Alibaba also had news closer to the ground: the company said its open-source Qwen-27B has become the most popular model among developers worldwide on desktop, and it launched Qwen Intelligence, an end-to-end solution that enables mobile devices to reason through and execute complex tasks.

Whether Alibaba can actually deliver a 5- to 10-trillion-parameter model — and the energy-hungry infrastructure to run it — remains to be seen. But with Tuesday's share rally adding to a strong year for the stock, the market is, for now, giving the company the benefit of the doubt. As Wu put it in closing his keynote, the more capable AI becomes, the more powerful humanity will be — the conviction he described as the ultimate purpose of Alibaba's commitment to the Machine Intelligence era.