Alibaba Unveils Zhenwu V900 AI Chip and Plans Qwen Model With Up to 10 Trillion Parameters in Full-Stack AI Push

Alibaba Group has escalated its push to build a self-sufficient artificial-intelligence ecosystem, unveiling a powerful new AI chip and outlining plans for a next-generation language model with between 5 trillion and 10 trillion parameters.

The announcements, made at the company's Apsara technology conference in Hangzhou and reported on 22 September 2026, laid out an increasingly ambitious strategy spanning semiconductors, foundation models, cloud infrastructure and data centres.

At the centre of the hardware announcement is the Zhenwu V900, a new AI chip that Alibaba says delivers roughly three times the performance of its predecessor and can be linked into large computing clusters. Mass production is expected to begin in early 2027.

A model of unprecedented scale

On the software side, Alibaba outlined plans for a next-generation model in its Qwen family with between 5 trillion and 10 trillion parameters. Parameters are the internal variables that a model learns during training, and while parameter count is not a direct measure of capability, a model at the upper end of that range would be among the largest ever publicly announced.

The Qwen family has become one of the most widely used series of open-weight models globally, adopted by developers and companies in China and abroad. Alibaba's decision to release many Qwen models openly has helped it build a large developer ecosystem, positioning Qwen as a key alternative to models from US companies.

Training a model of 5 trillion to 10 trillion parameters would require enormous computing resources, making the parallel investment in domestic chips and data-centre capacity essential to the plan.

Building the power base

Alibaba chief executive Eddie Wu said Alibaba Cloud intends to expand its data-centre capacity beyond 20 gigawatts by 2032. That figure underlines the scale of physical infrastructure now considered necessary to compete at the frontier of AI.

Data-centre capacity measured in gigawatts reflects the electricity required to power servers and cooling. The largest AI data-centre projects globally now plan for hundreds of megawatts or more at a single site, and the aggregate capacity required by leading AI companies has grown dramatically. Alibaba has previously pledged to invest more than RMB380 billion in cloud and AI infrastructure over three years, a commitment announced in early 2025.

The export-control backdrop

The announcements must be understood against the backdrop of US export controls, which have restricted Chinese companies' access to Nvidia's most advanced AI chips and to the equipment needed to manufacture leading-edge semiconductors. Those restrictions have accelerated efforts by Alibaba, Huawei and other Chinese companies to build alternatives to Nvidia hardware and the surrounding software ecosystem.

Alibaba's chip-design efforts, led by its semiconductor unit, have produced a series of processors for cloud and AI workloads. The Zhenwu V900 represents a significant step up in capability, and its ability to scale into large clusters is crucial: modern AI training relies not only on individual chip performance but on how efficiently thousands of chips can work together.

A credible domestic technology stack would give Chinese AI developers greater insulation from future semiconductor restrictions. It would also create another large-scale technology platform competing for developers, cloud customers and enterprise workloads, both in China and in markets across Asia, the Middle East and elsewhere.

The manufacturing question

Designing a competitive AI chip is one challenge; manufacturing it at scale is another. Advanced AI chips require cutting-edge fabrication processes, and Chinese companies have limited access to the most advanced manufacturing technology because of export restrictions on lithography equipment. Domestic foundries have made progress, but production capacity and yields at advanced nodes remain constraints.

The early-2027 mass-production timeline for the Zhenwu V900 will therefore be closely watched as an indicator of China's semiconductor manufacturing capabilities. Delays or limited volumes could constrain Alibaba's ability to deploy the chip across its data centres at the scale its ambitions require.

Competitive implications

For global AI competition, Alibaba's strategy highlights a trend towards vertical integration. The largest US technology companies, including Google, Amazon, Microsoft and Meta, have all invested in custom AI chips while continuing to buy Nvidia hardware. Alibaba is pursuing a similar path, but with greater urgency given the constraints it faces.

The full-stack approach offers advantages in cost control, performance optimisation and supply security. It also allows companies to design chips tailored to their own models and workloads. The trade-off is the enormous capital and engineering effort required to compete across every layer.

For Nvidia, which has seen its China business constrained by export rules, the emergence of increasingly capable domestic alternatives represents a long-term competitive challenge in one of the world's largest technology markets.

What it means for India and the Global South

For India and other emerging economies, Alibaba's announcements carry indirect but important implications. The global AI race is increasingly shaped by access to chips, energy and capital, and the competition between US and Chinese technology ecosystems will influence the options available to other countries. Open-weight models such as Qwen are already used by developers worldwide, including in India, as building blocks for applications.

India's own efforts, including its national AI mission, investments in compute capacity and initiatives to develop domestic foundation models, are unfolding in this broader context. The scale of Alibaba's plans underlines how much resource is now required to operate at the frontier.

Investors weigh the costs

Investors will be watching how Alibaba balances these ambitions with shareholder returns. Building chips, training frontier models and constructing gigawatt-scale data centres all require enormous capital expenditure, and the returns depend on the company's ability to monetise AI through cloud services, enterprise applications and consumer products. Alibaba Cloud's growth has become a key driver of investor sentiment towards the group, and sustained demand for AI computing will be essential to justify the spending.

The road ahead

Alibaba's Apsara announcements present a clear vision: a vertically integrated AI company with its own chips, frontier models and massive data-centre capacity. Execution risks are significant, from semiconductor manufacturing to the energy required for its data centres and the commercial returns on its investments.

Yet the direction of travel is unmistakable. As the AI race intensifies, the largest companies are no longer content to compete on software alone. Alibaba's latest moves signal that, in China at least, the future of AI will be built on home-grown silicon, and that the company intends to be at the centre of it.