Alibaba Unveils Zhenwu V900 as Qwen Roadmap Pushes Toward 10 Trillion Parameters

Alibaba has laid out one of its most ambitious artificial-intelligence roadmaps yet, combining plans for much larger Qwen models with a new in-house AI processor and a major expansion of cloud infrastructure.

At its Apsara Conference in Hangzhou on September 22, Alibaba said a future model could reach roughly 5 trillion to 10 trillion parameters. The company also introduced the Zhenwu V900, a processor developed by its T-Head semiconductor unit, and set a target for Alibaba Cloud’s global data-center capacity to exceed 20 gigawatts by 2032.

The announcements matter because Alibaba is not approaching AI as a single-product race. It is trying to control more of the stack: chips, data centers, cloud services, foundation models and the agents that ultimately use those models.

What Alibaba announced

Reuters reported that Alibaba’s next-generation ambitions include an AI model in the 5 trillion-to-10 trillion parameter range, potentially several times the scale of its current flagship. Parameter count alone does not determine model quality, but the figure signals the scale of training and infrastructure Alibaba expects to pursue.

The company also unveiled the Zhenwu V900. Alibaba says the processor can deliver about three times the performance of its predecessor and can be connected in large clusters designed for demanding AI workloads. Mass production is expected in early 2027.

Alibaba CEO Eddie Wu also set a goal for the company’s global data-center capacity to exceed 20 GW by 2032. That is an infrastructure target, not simply a model announcement, and it helps explain why Alibaba is investing simultaneously in silicon and cloud capacity.

Why the chip matters

The global AI boom has made advanced processors a strategic bottleneck. Training and serving increasingly capable models requires enormous amounts of compute, while access to the highest-end chips is affected by supply constraints, cost and trade restrictions.

For Alibaba, developing its own processors can reduce dependence on outside suppliers and give it more control over how its cloud systems are optimized. The economic question is not whether one chip can simply replace every alternative. It is whether Alibaba can make its hardware, networking, software and models work efficiently enough together to offer competitive AI services at scale.

That full-stack approach resembles a broader shift across the industry. AI companies increasingly compete not only on benchmark scores but also on inference cost, availability, power efficiency and how quickly new capacity can be deployed.

A 10-trillion-parameter model would be only part of the story

Very large parameter counts attract attention, but they should not be treated as a direct ranking of intelligence. Training data, architecture, post-training, inference-time reasoning and software optimization can all affect how useful a model becomes.

The more revealing signal is Alibaba’s willingness to plan infrastructure around models of this scale. Larger systems can demand more memory, networking capacity, electricity and cooling. If Alibaba follows through, the cost of supporting its AI roadmap will extend far beyond the training run itself.

This is similar to the infrastructure dynamic Earnyx has covered in Blackstone’s planned $25 billion Pennsylvania data-center project: the AI race is increasingly becoming a contest over physical capacity as well as software.

Alibaba is building from chips to agents

Alibaba’s strategic advantage is that it already operates a major cloud platform and a large digital-services ecosystem. A company that controls more layers can potentially tune its models to its own chips, distribute them through its cloud and embed them in commerce and enterprise services.

That does not guarantee success. Building competitive processors is difficult, advanced manufacturing remains concentrated among a small number of suppliers, and massive data centers require reliable power and cooling. Alibaba also faces strong domestic competition from other Chinese technology groups and increasingly capable open models.

Still, the roadmap shows that the company wants to compete at several layers simultaneously rather than rely on imported accelerators underneath its AI products.

The 20 GW target shows how capital-intensive AI is becoming

A 20-gigawatt global data-center target is significant because it translates AI ambition into electricity demand. Data-center capacity is constrained by grid connections, generation, transformers, cooling equipment, land and construction timelines. Even companies with capital cannot instantly add capacity.

Reuters reported that Wu described AI demand as exceptionally robust while acknowledging shortages across the data-center supply chain. That tension—strong demand but limited physical capacity—is becoming one of the defining economics of the AI industry.

It also creates opportunities beyond chipmakers. Cooling providers, utilities, electrical-equipment manufacturers, networking companies and construction firms all sit in the path between an AI model roadmap and usable computing capacity.

What to watch next

The most important next steps are execution rather than headline specifications. Watch whether the Zhenwu V900 reaches mass production on schedule, how broadly Alibaba deploys it inside Alibaba Cloud, and whether customers can use the resulting systems at competitive cost and performance.

It will also be important to see what Alibaba means in practice by a future 5-trillion-to-10-trillion-parameter model. The company has indicated that Qwen 4 is in training, but a longer-term model at the top of the stated range would represent another step in scale.

Finally, the 20 GW cloud target will be a useful measure of whether Alibaba can translate AI demand into physical infrastructure. Data-center projects can be delayed by power, equipment and permitting even when financing is available.

Bottom line

Alibaba’s September 22 announcements are bigger than a new chip or a larger Qwen model. They describe a strategy to build an increasingly self-contained AI platform spanning semiconductors, data centers, cloud computing, models and agents.

The headline 10-trillion-parameter ambition will draw attention, but the harder challenge is making all of those layers work together economically. If Alibaba can do that, the company could strengthen both its position in China’s AI market and its ability to compete for cloud and AI customers beyond China. If it cannot, the scale of the investment makes execution failures expensive. Either way, Alibaba has made clear that it sees the next phase of AI competition as a full-stack infrastructure race.