Alibaba has just made one of its boldest moves yet in artificial intelligence.

At its annual Apsara Conference in Hangzhou, the Chinese technology giant unveiled a new AI accelerator called the Zhenwu V900, announced plans for a next-generation AI model containing as many as 10 trillion parameters, and laid out a long-term expansion strategy that could take Alibaba Cloud’s data-center capacity beyond 20 gigawatts by 2032.

The announcement effectively combines three battles into one: AI chips, giant foundation models and computing infrastructure.

That is important because China’s AI ambitions increasingly depend on controlling the entire technology stack.

Alibaba is trying to build exactly that.

The headline hardware announcement was Alibaba’s new Zhenwu V900 AI chip, developed by the company’s T-Head semiconductor unit.

Alibaba says the V900 delivers about three times the performance of its predecessor, helped by improvements in GPU memory and bandwidth. The company says it can support both AI training and inference and can be assembled into very large clusters capable of supporting as many as 500,000 processing cards. Mass production is expected to begin in the first quarter of 2027.

Those numbers matter because AI development is increasingly determined not just by the performance of one accelerator, but by how thousands or hundreds of thousands of accelerators can work together.

The largest AI models cannot practically run on a single chip.

They require enormous clusters connected through high-speed networking and backed by equally large data-center infrastructure.

Alibaba is therefore attacking a much bigger target than traditional semiconductor competition.

It wants to supply the compute layer itself.

Alibaba also revealed plans for an AI model with between 5 trillion and 10 trillion parameters.

That would represent a huge increase over its current flagship Qwen 3.8 Max model, which has roughly 2.4 trillion parameters, according to reports about Tuesday’s announcement.

Parameter counts alone do not determine how intelligent an AI system is, and larger models are not automatically better.

But the scale of Alibaba’s target demonstrates the level of resources the company expects to devote to frontier AI.

A model with trillions of parameters requires extraordinary computing resources during training and potentially enormous infrastructure during inference.

That explains why Alibaba announced the hardware and data-center pieces alongside the model roadmap.

The company is effectively saying that it wants to build both the machine that trains the AI and the AI itself.

Alibaba CEO Eddie Wu has emphasized that the company intends to develop the full AI technology stack, including models, semiconductors and data centers.

That strategy has become more important as China faces restrictions on access to some of the world's most advanced foreign AI accelerators.

The United States continues to control exports of certain advanced semiconductor technologies, while Chinese technology companies have increasingly focused on developing domestic alternatives.

Alibaba cannot control every component of the global semiconductor supply chain.

But it can attempt to control more of the software and infrastructure around AI.

That includes chip design, model optimization, cloud deployment, networking and data-center capacity.

The Zhenwu V900 therefore has significance beyond its raw performance.

It represents Alibaba’s attempt to reduce dependence on external accelerator suppliers and create a platform that its own cloud and AI services can use at large scale.

Huawei has received enormous attention for its Ascend AI accelerator roadmap, and it remains one of the most important Chinese competitors in domestic AI computing.

But Alibaba’s entry makes the competitive landscape broader.

China now has multiple major technology groups working on advanced AI hardware, including Alibaba’s T-Head, Huawei and semiconductor specialists such as Cambricon, while cloud providers and model developers continue building alternative architectures.

That diversification can be strategically important.

Instead of relying on a single national champion, China’s technology ecosystem can experiment with different chip designs, software stacks and deployment models.

Alibaba can also use its enormous cloud infrastructure as a testing ground.

That is an advantage pure chip startups do not necessarily possess.

Alibaba has committed more than $53 billion over three years toward integrating AI into its businesses, according to The Wall Street Journal.

That money is intended to support AI across Alibaba’s operations, including its cloud infrastructure and e-commerce ecosystem.

This is crucial because Alibaba is not trying to become a chip company for the sake of selling chips.

Its hardware investments are tied to the company’s broader AI strategy.

The more AI workloads Alibaba runs internally, the greater the incentive to optimize its own hardware and software stack.

And the more AI customers Alibaba Cloud attracts, the more valuable that infrastructure becomes.

The strategy resembles the vertically integrated approach increasingly seen elsewhere in the global AI industry.

Instead of purchasing every layer from different vendors, the largest technology companies increasingly want control over the hardware, networking, software and cloud environment surrounding their models.

The company recently raised approximately $10.2 billion through a Hong Kong share offering, with AI investment among the major purposes for the capital, according to reporting surrounding the new strategy.

The market appears to have welcomed the announcements.

Alibaba shares rose roughly 5.1% on Tuesday, reaching a one-month high, as investors reacted to the chip launch, model plans and broader AI strategy.

That reaction suggests investors are increasingly treating AI infrastructure as a central component of Alibaba’s future rather than simply another cloud-business initiative.

But a giant AI strategy also brings giant costs.

Alibaba says it wants its cloud data-center capacity to exceed 20 gigawatts by 2032.

That figure illustrates how quickly the economics of computing are changing.

Traditional software companies could scale largely by copying software.

AI infrastructure does not work that way.

Every major expansion requires electricity, cooling, networking, land, buildings and semiconductor equipment.

The power requirement alone can become a strategic issue.

A 20-gigawatt portfolio represents a vast industrial infrastructure commitment, and Alibaba will ultimately have to balance utilization, capital expenditure and customer demand.

That is one reason the chip strategy and cloud strategy are inseparable.

If Alibaba builds more efficient accelerators, it can potentially increase the amount of AI work performed per unit of electricity and infrastructure.

The new chip is undoubtedly important, but performance claims need careful interpretation.

Alibaba says the V900 is three times faster than its predecessor. That is a company claim, and real-world competitiveness depends on workload, software optimization, memory bandwidth, networking and the entire cluster architecture.

The same principle applies to the 5 trillion-to-10 trillion parameter model.

A larger parameter count does not automatically mean a better AI experience.

Competitors can outperform larger models through better training data, architecture, reinforcement learning, inference optimization or software.

Alibaba therefore has another difficult task beyond designing powerful hardware: making its chips and models work together efficiently enough that customers actually prefer the complete system.

That is where its cloud business could become a decisive advantage.

The company's strategy also reflects a broader change in the AI market.

AI systems are beginning to perform increasingly complex tasks, shifting demand from simple question-answering toward reasoning, coding, content creation, search, enterprise automation and autonomous agents.

Those workloads require more computation.

They also create demand for hardware designed around inference, not only model training.

Alibaba’s Zhenwu V900 is explicitly designed to support both training and inference, reflecting that broader market shift.

The company is therefore positioning itself for a future where AI applications run continuously rather than being occasional experiments.

That could create a huge market for cloud computing.

The most important message from Alibaba’s announcement may not be the 10 trillion-parameter target or even the new chip.

It may be the decision to combine them.

Alibaba is attempting to create an integrated technology chain stretching from semiconductor design to data centers to foundation models to cloud services and consumer applications.

That strategy reflects an increasingly important reality in AI: control of one layer is valuable, but control of multiple layers can potentially produce much tighter optimization and greater strategic independence.

The result is a much more competitive Chinese AI landscape.

Alibaba is expanding chips.

Huawei is accelerating its accelerator roadmap.

Tencent is advancing foundation models and image-generation systems.

Other Chinese companies are developing specialized models and hardware.

And all of them are operating in an environment where computing capacity has become a strategic resource.

The Zhenwu V900 is therefore more than Alibaba’s newest AI chip.

It is a symbol of the company’s attempt to turn itself into an integrated AI infrastructure giant — one capable of designing the hardware, training the models and operating the data centers needed to run them.

The next test will be execution: whether the V900 can reach mass production, whether Alibaba’s enormous model ambitions translate into practical AI products, and whether its expanding infrastructure can generate returns large enough to justify the scale of the investment.

For now, Alibaba has made its intention clear: it does not want merely to use the AI revolution.

It wants to build more of the machinery behind it.