Hangzhou: Alibaba has unveiled a broad artificial-intelligence roadmap spanning proprietary chips, cloud infrastructure, foundation models and mobile agents. The centrepiece is the Zhenwu V900, which the company describes as China’s most powerful AI chip, alongside confirmation that its next-generation Qwen 4 model is in training.
The announcements were made at Alibaba Cloud’s annual Apsara Conference on September 22. According to the company, the V900 delivers about three times the performance of its previous Zhenwu M890 chip. Alibaba also said future Qwen models could scale to between five trillion and 10 trillion parameters, compared with 2.4 trillion for its current Qwen3.8-Max model.
What Alibaba announced
The roadmap covers more than a single processor. Alibaba is building an integrated stack that includes AI accelerators, servers, high-performance networking, distributed storage, model-training services and agent platforms. The strategy resembles the vertically integrated approach pursued by the world’s largest cloud providers: own more of the hardware and software layers so performance, cost and deployment can be managed together.
Zhenwu chips are intended for Alibaba’s data centres and cloud clients. They support both training, where a model learns from large datasets, and inference, where a trained model produces answers or predictions. In practical terms, inference capacity has become just as important as training because widely used AI products must respond to millions of requests at predictable cost.
Qwen 4 and the parameter race
Alibaba said Qwen 4 is now in training, while outlining a future scale of up to 10 trillion parameters. Parameter count is one measure of model capacity, but it is not a complete measure of quality. Data selection, architecture, training efficiency, reasoning methods and post-training can matter as much as raw size.
The company highlighted an experiment in which a model made more than 10,000 calls to electronic-design automation tools during over 60 hours of self-improvement, producing production-grade chip bus modules. Because this is a company claim, independent testing will be needed before drawing broad conclusions about engineering productivity.
Why the chip matters for China
Chinese technology groups have accelerated work on domestic AI hardware as US-led export controls restrict access to some advanced chips and manufacturing equipment. Alibaba’s announcement follows Huawei’s Atlas 960E AI SuperPoD launch and adds another major competitor to China’s domestic compute ecosystem.
The challenge is not only designing a fast chip. Commercial success depends on manufacturing yield, memory bandwidth, software tools, power efficiency and a developer ecosystem. Nvidia’s strength, for example, comes from its CUDA software platform as well as its processors. Alibaba will need to show that customers can move workloads onto Zhenwu hardware without costly rewrites or reliability trade-offs.
Cloud demand is the business case
Alibaba said demand for AI computing is rising exponentially and that it will expand the data centres supporting its cloud business. That matters because chip development requires large, recurring demand to justify investment. Alibaba can serve as both supplier and anchor customer, using its own e-commerce, cloud and AI services to test the platform at scale.
The launch also arrives before high-level US-China talks where AI, trade and tariffs are expected to feature. Our report on Xi Jinping’s US visit explains the diplomatic setting, while US-China AI safety discussions show that competition and risk management are developing in parallel.
What remains unverified
Alibaba’s performance figures and “most powerful in China” description are vendor claims. Independent benchmarks, production volumes, energy use and customer deployment data have not yet established how the V900 compares with rival accelerators. The planned size of Qwen 4 also describes a training roadmap, not a released model.
Still, the breadth of the announcement is strategically important. Alibaba is signalling that it wants to compete across the complete AI supply chain rather than depend on outside hardware. If execution matches the roadmap, the company could strengthen China’s ability to build and operate advanced AI systems under tighter technology controls.
Sources / References
- Alibaba Cloud: full-stack AI roadmap, 22 September 2026
- Associated Press: Alibaba chip and model announcement
The Tech Desk covers technology, artificial intelligence and cybersecurity. Stories are checked against at least two credible sources before publication.
