Alibaba unveils Zhenwu V900 AI chip, plans 10T parameter model

Alibaba Unveils Zhenwu V900 AI Chip

Alibaba Group Holding unveiled its Zhenwu V900 artificial intelligence chip on Tuesday at its annual Apsara conference in Hangzhou, China. The company described the processor as China's most powerful AI chip, delivering three times the performance of its Zhenwu M890 predecessor.

Alibaba also disclosed plans to train an artificial intelligence model with up to ten trillion parameters in pursuit of superintelligence. Alibaba shares rose about three percent in Hong Kong trading following the chip launch and its expanded data center plans.

Chairman Joe Tsai said Alibaba remains committed to building comprehensive artificial intelligence capabilities spanning chips, cloud infrastructure and advanced models. Chief Executive Officer Eddie Wu Yongming called AI models, AI chips and AI cloud the three core foundations of machine intelligence.

Alibaba Cloud, the group's computing unit, aims to expand its global data center capacity to more than twenty gigawatts by the year 2032, company executives said.

The Apsara conference, held each year in the eastern Chinese city of Hangzhou, serves as Alibaba's main platform for announcing new cloud and artificial intelligence products.

Ambition Targets Superintelligence and Market Control

Alibaba's announcement today underscores its broader ambition to secure still tighter control over the strategic computing infrastructure that powers the world's most advanced artificial intelligence systems.

Alibaba said it plans to train Qwen 5, its next flagship model, with between five trillion and ten trillion parameters. The largest model currently in the Qwen family, Qwen3.8-Max, contains roughly 2.4 trillion parameters and remains the group's most capable system.

Liu Dayiheng heads the Qwen large language model project at Alibaba Token Hub and announced the roadmap for future Qwen releases. He highlighted advances in recursive self-improvement driven by empirical feedback and upgrades to speech, image and multimodal data processing at Alibaba.

Alibaba also launched Qwen Intelligence, an agent platform offering smartphone manufacturers tools built on its Qwen models, and confirmed earlier work with Apple on Chinese iPhones.

Despite United States export controls limiting China's access to advanced chips, the country's leading artificial intelligence models are still largely trained on Nvidia processors, media reports said.

Chip Specs and Production Timeline Revealed

Gao Hui, a vice president at Alibaba's chip unit T-Head, said the Zhenwu V900 will carry 216 gigabytes of high-bandwidth memory. Bandwidth between chips will reach 1.2 terabytes per second, and clusters can connect up to 500,000 processors for AI training.

Mass production and commercial release of the new Zhenwu V900 chip are scheduled to begin in the first quarter of 2027, and annual supply will rise sharply.

T-Head also revealed a next-generation Zhenwu J900 processor expected to launch in the third quarter of 2028 for artificial intelligence workloads. The unit plans proprietary central processing units for agentic AI tasks in 2027, including the Yitian 720 and Yitian 730 models.

The Yitian 730 will be T-Head's first central processing unit built on a microarchitecture developed internally by the company itself. Current Zhenwu chips already serve more than 650 customers across automotive, financial, energy and manufacturing sectors, according to the company.

The intensifying domestic competition includes Huawei, which unveiled its new Atlas 960 SuperPoD computing cluster at a major annual conference in Shanghai, China, on September 17.