Alibaba unveiled its self-developed AI chip, Zhenyu V900, and set out a plan to expand Alibaba Cloud's global data center capacity to 20 GW by 2032. It aims to challenge an Nvidia-led AI computing market by building an ecosystem spanning in-house semiconductors, cloud infrastructure and AI models.

On Sept. 22 local time, blockchain outlet Cryptopolitan reported that Alibaba unveiled the Zhenyu V900 and said it targets top-tier performance in China.

Eddie Wu (우딩), Alibaba's chief executive officer, said at the company's annual Apsara Conference that Zhenyu V900 performance has improved threefold from the previous-generation Zhenyu M890. He also said it can connect up to 500,000 chips for advanced AI training.

The announcement goes beyond introducing a new AI chip. Alibaba is strengthening AI chip capabilities around its in-house chip design unit T-Head, while also accelerating expansion of cloud data centres.

Alibaba plans to expand global data centre capacity to 20 GW by 2032. It aims to link Zhenyu V900, cloud computing infrastructure, in-house foundation models and AI agents into a single ecosystem.

Infrastructure investment is also sizeable. Alibaba has already approved a plan to invest more than $53 billion over the next three years in AI and cloud infrastructure. Part of the funding was raised through a $10.2 billion share sale in Hong Kong in August.

Wu said at the Apsara Conference that customer demand related to AI is rising sharply, and presented infrastructure expansion as a new core business area.

Alibaba is also expanding its AI models and services. It unveiled the Qwen family of foundation models, multimodal models, an agent-based cloud and a mobile AI agent platform. Alibaba Group Chairman Joe Tsai said AI can be deployed in various real-world scenarios and can raise productivity across industries.

Market reaction was immediate. Alibaba shares rose more than 5 percent in Hong Kong trading, and Tencent also gained more than 7 percent. Some in the market also speculated that Alibaba is considering training a large AI model with 5 trillion to 10 trillion parameters.

The move comes as U.S. restrictions on semiconductor exports to China continue, and as Chinese big tech companies seek to secure their own AI computing ecosystems. With China's access to cutting-edge Nvidia AI chips constrained, securing in-house AI semiconductors and data centres has emerged as a key task for Chinese companies.

U.S.-China technology friction around Alibaba is also continuing. U.S. authorities recently alleged that Alibaba and Chinese AI company DeepSeek improperly used knowledge from major U.S. AI models to advance their own technology. China denied this and said it would respond if the United States uses the allegations as a means to pressure Chinese companies.

The Federal Bureau of Investigation, National Security Agency and the Cybersecurity and Infrastructure Security Agency also alleged that Chinese technology companies have systematically collected major U.S. AI models for more than 2 years. The agencies said proprietary data from ChatGPT, Gemini, Claude and Grok was used to build Chinese AI systems.

China's Foreign Ministry, meanwhile, said all parties should strengthen cooperation to promote open and inclusive AI development.

With AI set to be a key agenda item at a U.S.-China summit, Alibaba's announcement is drawing attention as a large-scale investment plan unveiled at a time when China's AI self-reliance strategy intersects with U.S.-China tech rivalry.

Key questions going forward are Zhenyu V900's real-world performance, mass production capacity, and how efficiently Alibaba can connect it with a 20-GW cloud infrastructure.