Alibaba unveils Zhenwu V900 AI chip, targets 20 GW cloud buildout by 2032

Shares of Alibaba jumped roughly 3% in Hong Kong on Tuesday after the Chinese technology group unveiled a next-generation AI chip and a roadmap to expand its cloud data center capacity to more than 20 gigawatts by 2032.

At Alibaba Cloud's annual Apsara Conference in Hangzhou, the company introduced the Zhenwu V900, an AI accelerator it claims delivers three times the performance of its predecessor, the Zhenwu M890 released in May. Mass production and commercial release of the V900 are scheduled for the first quarter of 2027.

The announcements land as Alibaba's Western peers spend aggressively on AI infrastructure:

• Earlier this month, Nvidia outlined plans to support up to 2 gigawatts of AI compute in Australia by 2027 alongside local partners.
• Meta in July said it would build a 1-gigawatt data center in Alberta, Canada, a project expected to cost roughly $9 billion and take up to three years to complete.

What did Alibaba actually announce?

Alibaba split the news into three layers: silicon, data centers, and models.

The Zhenwu V900 is the centerpiece. Existing Zhenwu chips are already deployed at more than 650 customers across automotive, finance, energy, and manufacturing, the company said. The V900 is meant to scale that footprint, with a production date locked to Q1 2027.

On capacity, Alibaba Cloud said it would grow its global data center footprint past 20 GW by 2032. That figure dwarfs the 2 GW Australian commitment Nvidia discussed and the 1-gigawatt Alberta facility Meta plans.

On models, Alibaba confirmed that its next-generation Qwen 4 model is in training. It also outlined plans for Qwen 4.5 and Qwen 5 series successors, though it released no training data, parameter counts, or release dates.

How does the Zhenwu V900 compare to its predecessor?

The Zhenwu M890 launched in May and is the chip the V900 will replace. Alibaba described the V900 as delivering three times the performance of the M890. The company did not publish the benchmark methodology, the model class used for measurement, or the precision format.

By reserving the V900 for Q1 2027, Alibaba buys itself roughly two years to fill orders for the existing M890. The new chip will arrive in a market where Western alternatives remain constrained for Chinese customers.

The 650-customer base spans four named verticals. Alibaba did not detail the share of revenue from each. The chip is already running in production environments rather than pilot programs.

Why does the 20 GW target matter?

Twenty gigawatts approaches the continuous output of twenty large nuclear reactors. It is also roughly twenty times the single-site capacity Meta is building in Alberta.

Alibaba did not disclose:

• Which regions the new capacity will land in
• Capital expenditure figures
• How much of the buildout depends on in-house Zhenwu silicon versus third-party GPUs
• Power purchase agreements or grid partners

The 2032 horizon is also long. By then, three generations of Qwen models will have shipped, and the V900 will be mid-life. The roadmap is a statement of ambition, not a quarterly forecast.

For investors, the figure serves as a marker of Alibaba's seriousness. Building 20 GW of data center capacity requires land, power, cooling, and silicon supply chains. Alibaba has signaled it intends to secure all four.

How does this fit the broader AI buildout?

China's cloud and chip vendors are racing to match Western hyperscalers on raw compute. Alibaba's domestic peer Huawei unveiled AI infrastructure last week that it said can scale to as many as one million processors.

Alibaba's pitch is vertical: it designs the chip, builds the data center, trains the model, and sells the cloud service. Western peers split those layers across separate companies.

The vertical model has advantages. Coordination between silicon and software is tighter. Capital decisions can be made in one boardroom. The disadvantage is concentration risk if any single layer underperforms, and the capital burden is heavier on a single balance sheet.

What did Alibaba's leadership say?

Eddie Wu, CEO of Alibaba Group, framed AI as still in its infancy. Machine thinking has an "enormous growth runway," Wu said at the conference. He compared today's AI development with the early stages of electrification, adding: "AI coding is simply the light bulb of the machine intelligence era."

The metaphor matters. Wu is positioning Alibaba for a multi-decade buildout, not a quarterly revenue pop. Investors responded in kind. The 3% Hong Kong move came on volume, not just headlines.

What about the model roadmap?

Qwen has become Alibaba's most visible AI product line. The company confirmed that Qwen 4 is in training and laid out plans for Qwen 4.5 and Qwen 5 series as successors.

The release cadence will determine how Alibaba competes as the global frontier moves forward. Alibaba gave no release dates and no parameter or training-compute figures for any of the three models.

What is clear is that Alibaba intends to keep shipping models at a steady cadence. The roadmap slide shows three named model lines over a multi-year window, a deliberate signal that the company plans to stay in the frontier race through the end of the decade.

What questions remain?

Three will shape the next twelve months:

1. Can Zhenwu V900 hit production in Q1 2027 on schedule?
2. Will Alibaba disclose which regions absorb the 20 GW?
3. When does Qwen 4 ship, and which frontier models will it be benchmarked against?

Alibaba did not address any of these on Tuesday. Investors, judging by the 3% move, focused instead on the headline target.

The stock move signals that the market read the announcements as credible direction. Production timelines, regional splits, and model results will determine whether that confidence holds through 2026 and into the V900 launch year.