What Is Alibaba’s Zhenwu V900 and How Does It Support the Qwen AI Roadmap?
Alibaba is building its next AI push from both ends. At the Apsara Conference in Hangzhou on September 22, CEO Eddie Wu unveiled a new accelerator for training and inference while laying out a roadmap for Qwen models that could eventually reach 10 trillion parameters.
“At the conference, Alibaba revealed that its next-generation model, Qwen 4, is currently in training. The company further announced its roadmap for the upcoming Qwen 4.5 and Qwen 5 model series, projected to scale up to 5 to 10 trillion parameters,” Alibaba said in its Apsara Conference announcement.
The hardware is the Zhenwu V900, developed by Alibaba’s T-Head chip division. “The Zhenwu V900 delivers three times the performance of its predecessor, the Zhenwu M890,” Alibaba said, adding that the accelerator has 216GB of GPU memory and 1,200GB/s of inter-chip bandwidth.
The model roadmap goes much further out. Alibaba says Qwen 4 is currently in training, while the following Qwen 4.5 and Qwen 5 generations are projected to scale to between 5 trillion and 10 trillion parameters. That would be up to roughly four times the 2.4 trillion parameters in the company’s current Qwen3.8-Max flagship.
The Chip Behind The Plan
V900 is designed as a unified training and inference processor rather than a chip aimed at only one stage of AI development.
Its support for FP8 and FP4 precision is intended to improve efficiency for different workloads, while Alibaba says the accelerator can be combined into a supernode cluster containing as many as 500,000 cards. The accompanying server design combines the V900 with Alibaba’s ICN Switch, Panmai SmartNIC and Zhenyue storage controller.
The model roadmap is moving into territory where a single accelerator is irrelevant. Training and serving models with trillions of parameters requires thousands of processors to communicate efficiently, along with enough memory, networking and storage to keep those processors supplied with data.
“By significantly reducing inference costs while increasing compute density, [the V900] seamlessly handles both high-precision model training and ultra-low-precision inference,” Alibaba said in its announcement.
Qwen Is Going Much Larger
Alibaba says Qwen 4 is already in training, but it has not announced a public release date, pricing or performance benchmarks for the model. The 5 trillion-to-10 trillion parameter target applies to Qwen 4.5 and Qwen 5, not the Qwen 4 generation itself.
Parameters are a measure of model size, not a direct score for intelligence. A larger model can still perform worse on particular tasks depending on its architecture, training data, optimization and inference efficiency.
“Vast numbers of AI agents are poised to take on an ever-greater share of work in the digital economy, each powered by tokens generated from models. Agents will increasingly serve as the primary interface between humans and the digital world,” said Eddie Wu, CEO of Alibaba Group, in a letter to shareholders.
Chips In House, Power On Tap
Alibaba Cloud says it wants the global data-center capacity it operates to exceed 20 gigawatts by 2032. The company also unveiled the Yitian 720 and Yitian 730 server CPU roadmap for 2027, extending its in-house hardware effort beyond AI accelerators.
“With our full-stack AI strategy, we have put Alibaba in a superior position to capture the substantial growth of demand for artificial intelligence and AI compute,” said Eddie Wu, CEO of Alibaba Group.
The strategy also reflects the constraints facing Chinese AI companies. Huawei accelerated its own next-generation Ascend chip roadmap at its September 17 conference, while U.S. restrictions have limited Chinese companies’ access to some of Nvidia’s most advanced processors.
For Alibaba, building more of the stack internally reduces the number of points at which its AI plans depend on an outside supplier. It does not eliminate the manufacturing constraints facing China’s semiconductor industry, but it gives the company more control over the hardware and infrastructure it can deploy domestically.
The Stack Must Work Together
Alibaba’s more consequential bet is that its own silicon, Qwen models and cloud infrastructure can be designed to work together efficiently enough to support increasingly large AI systems. If that works, the company gains more control over the economics and availability of its AI stack.
That remains unproven as the V900 does not enter commercial release until Q1 2027, Qwen 4 is still in training, and the 5-trillion-to-10-trillion target belongs to future model generations. Manufacturing capacity, real-world chip performance and the eventual quality of those models will determine whether the strategy works.
Alibaba has made the direction clear: rather than building one domestic alternative to Nvidia, it is trying to build enough of the chip, model and infrastructure stack itself that its AI ambitions depend less on what it can obtain from outside China.
