Alibaba Unveils Zhenwu V900 AI Chip and Plans Massive 20GW Data Center Expansion by 2032

Alibaba Launches Zhenwu V900 AI Chip with Triple Performance

Alibaba CEO Eddie Wu unveiled the Zhenwu V900 AI chip at the company's Apsara Conference in Hangzhou, positioning it as China's most powerful AI chip1

1

2

2

. The processor delivers three times the performance of its predecessor, the Zhenwu M890, which launched in May4

4

. Developed by Alibaba's T-Head semiconductor business, the Zhenwu V900 AI chip features 216GB of memory and 1200 GB/s inter-chip interconnect bandwidth1

1

. The chip supports formats from FP32 through FP4 and includes an improved Tensor Core arithmetic unit optimized for frontier model training and inference workloads1

1

. Alibaba can build single clusters packing up to 500,000 V900 units to power frontier AI model development5

5

. The new chip enters mass production and commercial release in the first quarter of 2027, while existing Zhenwu chips already serve more than 650 customers across automotive, finance, energy, and manufacturing sectors4

4

.

Alibaba Plans 20GW Data Center Expansion Through 2032

Alibaba announced plans to expand Alibaba Cloud's global data center capacity to over 20GW by 20322

2

3

3

. Eddie Wu cited exponentially rising demand for AI computing capacity as the driver behind this expansion3

3

. However, this timeline appears measured compared to competitors. Commercial real estate firm Cushman and Wakefield reported 37.7GW of datacenters currently under construction in the USA alone1

1

. The Stargate project promises 10GW online by 2029, while Meta announced plans for a 5GW datacenter campus1

1

. Amazon added 3.8GW of capacity in just twelve months during late 20251

1

. Wu acknowledged that global shortages across the AI infrastructure supply chain currently limit scaling speed3

3

. Alibaba recently launched its first Latin American cloud region in São Paulo, Brazil, featuring two data centers offering AI-related cloud services4

4

.

Ambitious AI Model Development Targets 10 Trillion Parameters

Alibaba plans to train a new AI model with 5 trillion to 10 trillion parameters2

2

5

5

. The company has started training its next-generation model, Qwen 4, with two successors on its roadmap expected to scale to these parameter counts1

1

. This represents a significant leap from its current Qwen3.8-Max model, which has 2.4 trillion parameters3

3

. Alibaba's Qwen team is exploring Recursive Self-Improvement (RSI), technology where models design new models, and has made meaningful progress in this area1

1

. The company is regarded as one of China's foremost AI developers, having invested tens of billions of dollars into developing its flagship Qwen models5

5

. Earlier this month, Alibaba expanded its enterprise AI tools with an upgraded QoderWake platform that creates AI-powered digital employees capable of working across multiple platforms, with nearly 100,000 digital employees deployed since April completing about 2 million tasks in three months4

4

.

Strategic Context Amid Geopolitical Tensions and Export Restrictions

The announcement comes days before Chinese leader Xi Jinping's state visit to Washington for meetings with U.S. President Donald Trump, where AI competition is expected to be a major theme3

3

. U.S. AI leaders, including Anthropic CEO Dario Amodei, have warned that China's AI development poses threats to the United States3

3

. Recent Chinese advancements in AI are giving Beijing more leverage in upcoming talks with Washington, as China steps up its technological self-sufficiency push despite U.S.-led export restrictions barring access to cutting-edge AI chips and chipmaking machines3

3

. Alibaba has invested heavily in developing its own AI accelerator processors as part of Beijing's broader effort to foster home-grown AI chips5

5

. Open Chinese models, often more affordable than cutting-edge closed models from leading U.S. AI labs, have been making inroads globally, including in the U.S.3

3

.

Vision for Machine Intelligence Dominance

Eddie Wu likened current AI applications to light bulbs during the industrial revolution, suggesting the really important applications will emerge decades later1

1

. He compared AI to steam engines that initially merely replicated what horses and laborers did but eventually came to drive 99.9% of the world's physical work1

1

. Wu predicts that "machine intelligence" will one day perform 99.9% of all cognition, up from less than 3% currently, eventually producing over 1,000 times more thinking than all of humanity combined3

3

. He envisions AI breaking down ultra-complex tasks like building a Mars-capable starship into tens of millions of subtasks executed by millions of agents working non-stop, with humans only needing to define intent and goals1

1

. Alibaba's Hong Kong shares surged 4% following the announcement, while U.S.-listed shares closed 2.19% higher at $115.75 and rose another 1.12% after-hours4

4

5

5

.