HANGZHOU, China | September 22, 2026 —
Alibaba has unveiled one of its most ambitious artificial intelligence roadmaps yet, laying out plans for an AI model containing as many as 10 trillion parameters, a powerful new homegrown AI chip and a massive expansion of its global cloud infrastructure.
Alibaba CEO Eddie Wu announced the strategy at the company’s 2026 Apsara Conference in Hangzhou, positioning the Chinese technology giant for competition across the entire AI stack — from models and processors to data centers.
Investors responded quickly. Alibaba shares jumped around 5% following the announcements.
Alibaba targets a 5-to-10-trillion-parameter AI model
The headline announcement centers on the next generation of Alibaba’s Qwen AI family.
The company plans to train a future model containing between 5 trillion and 10 trillion parameters.
That would make it roughly two to four times larger by total parameter count than Alibaba’s current flagship, Qwen3.8-Max, which contains 2.4 trillion parameters.
Alibaba says the larger model will target increasingly difficult and long-duration tasks.
That could include AI agents capable of planning, reasoning and executing complicated projects over much longer periods with less human intervention.
Qwen3.8-Max already has 2.4 trillion parameters
Alibaba launched Qwen3.8-Max in August.
The company describes it as a sparse Mixture-of-Experts model containing 2.4 trillion total parameters, while activating a much smaller subset during individual tasks.
It can work with text, images and video and has been designed for coding, professional workflows and long-horizon operations.
The scale now proposed for Alibaba’s future models represents another major jump.
However, parameter count alone does not determine how intelligent or capable an AI model will be. Architecture, training data, compute efficiency and post-training techniques can matter just as much.
Alibaba also unveils Zhenwu V900 AI chip
Models are only one part of Alibaba’s plan.
Its semiconductor unit, T-Head, has unveiled the Zhenwu V900, a new AI accelerator designed to support training and inference for frontier models.
Alibaba says the V900 can deliver around three times the performance of its predecessor, the Zhenwu M890.
CEO Eddie Wu described it as the most powerful AI chip currently developed in China.
That remains Alibaba’s characterization rather than an independently established industry ranking.
Up to 500,000 chips in a cluster
The scale of the infrastructure plan is particularly striking.
Alibaba says Zhenwu V900 accelerators can eventually be connected in clusters containing as many as 500,000 chips.
Such large computing systems are increasingly important because frontier AI models require enormous amounts of processing power during training.
As model sizes increase from billions to trillions of parameters, the infrastructure behind them becomes as important as the software itself.
The V900 is expected to play a central role in Alibaba’s next generation of AI computing systems.
Commercial rollout targeted for 2027
Systems based on the Zhenwu V900 are expected to move toward commercial availability in the first quarter of 2027.
The new processor follows the Zhenwu M890, which Alibaba introduced earlier in 2026 as part of its effort to build more of its own AI computing technology.
Developing proprietary chips gives Alibaba greater control over performance, cost and supply.
It also has strategic importance as access to some advanced US-designed AI processors remains constrained in China.
Alibaba wants more than 20GW of global data-center capacity
Alibaba is not stopping at models and processors.
The company has set a target for Alibaba Cloud’s global data-center capacity to exceed 20 gigawatts by 2032.
That is an enormous infrastructure commitment.
AI data centers require electricity not just to operate processors but also to power networking systems, storage and cooling equipment.
The target shows how strongly Alibaba expects global demand for AI computing to expand over the coming years.
Alibaba is building the entire AI stack
The announcements reveal the broader strategy.
Alibaba wants to control multiple layers of its AI ecosystem:
Models: the Qwen family.
Chips: T-Head’s Zhenwu processors.
Cloud: Alibaba Cloud infrastructure.
Data centers: dramatically expanded global computing capacity.
Bringing those pieces together could allow Alibaba to optimize its systems from silicon to software rather than depend entirely on outside suppliers.
Why US chip restrictions matter
The strategy also reflects the changing semiconductor landscape.
Restrictions affecting the export of advanced American AI processors to China have increased pressure on Chinese technology companies to develop domestic alternatives.
Alibaba has consequently invested heavily in its own processor designs.
The Zhenwu V900 does not automatically mean Alibaba has matched Nvidia’s most advanced chips. Direct performance comparisons require independent benchmarks using equivalent workloads.
Nevertheless, Alibaba’s ability to design models, processors and cloud systems internally reduces its dependence on foreign technology.
Alibaba shares jump after announcements
Investors reacted positively to the Apsara Conference announcements.
Alibaba’s shares rose about 5%, reflecting enthusiasm around the company’s expanding AI strategy.
The move also highlights how important artificial intelligence has become to Alibaba’s broader growth story.
What began primarily as an e-commerce empire is increasingly positioning itself as an AI and cloud infrastructure company.
Alibaba’s biggest AI bet yet
The numbers tell the story.
A model targeting up to 10 trillion parameters.
A new processor claiming three times the performance of its predecessor.
Clusters potentially containing 500,000 accelerators.
And more than 20 gigawatts of data-center capacity by 2032.
Together, these announcements show that Alibaba is no longer treating AI as another product category.
It is attempting to build an entire computing ecosystem around it.
The next test will be execution: whether Alibaba can turn its enormous model, chip and infrastructure ambitions into systems that deliver competitive real-world performance at scale.
