The most powerful AI chip in China today, which offers three times the performance of its predecessor, the Zhenwu M890, is the Zhenwu V900. Eddie Wu, Alibaba’s CEO, announced the processor at the company’s annual cloud conference, Apsara, in Hangzhou. The chip, developed by T-Head, Alibaba’s chip design unit, is designed for high-precision model training and ultra-low-precision prediction.
See also: 18 malicious npm packages deliver Cross-Platform RAT to Alibaba tools users
T-Head has published a basic spec sheet for the chip, which includes 216GB of memory and 1,200 GB/s of inter-chip interconnect bandwidth. The processor supports FP32 to FP4 natively, with improvements made to the Tensor Core arithmetic unit for better instruction accuracy at FP8 and FP4. A single cluster can house up to 500,000 of these chips, enabling advanced model training and prediction.
Mass production and commercial launch of the V900 is scheduled for the first quarter of 2027. Wu expects a significant increase in annual AI chip shipments, with plans to refresh the series every year. Shipments of the Zhenwu family are forecast at 560,000 units to over 400 external customers, with Alibaba reporting more than 650 customers across various industries.
The Zhenwu V900 represents a significant step forward in AI chip technology, as its increased performance enables the execution of more complex and demanding computational tasks. Support for multiple precision levels, from FP32 to FP4, makes the chip highly versatile for a variety of applications, from training large models to low-precision prediction. This flexibility is critical for developing AI that can adapt to different environments and requirements.
Additionally, Alibaba unveiled a roadmap for its next-generation Yitian server processors, designed for agentic AI, which are expected to be released in 2027. The company is also training Qwen 4, with plans for Qwen 4.5 and Qwen 5, which are expected to scale to between five and 10 trillion parameters. The current top model, Qwen 3.8 Max, has 2.4 trillion parameters.
See also: Qwen3.8-Max: Alibaba challenges OpenAI and Anthropic
Wu stressed that the development of larger models aims to achieve artificial superintelligence, aiming to perform longer and more complex tasks. The Qwen team is also working on self-improvement through iteration, where a model identifies its weaknesses and generates training data with minimal human intervention. This approach can accelerate the evolution of artificial intelligence, making it more autonomous and efficient.
On the infrastructure front, Wu set a goal for Alibaba Cloud to reach 20GW of global data center capacity by 2032. Citigroup estimates that this capacity could generate over $160 billion in external revenue for the cloud division. Wu noted that global shortages in the AI data center supply chain are currently limiting Alibaba’s ability to scale its computing resources.
Alibaba Cloud plans to bring its AI hypernodes to commercial scale this quarter, although the locations of the new data centers have not been disclosed. The company has committed more than $53 billion over three years to expand AI and raised about $10.2 billion in a stock sale in August.
See also: ByteDance and Alibaba halt customized AI conversations in China
Alibaba’s development of the Zhenwu V900 and related technologies is part of a broader strategy to strengthen the company’s presence in the global artificial intelligence market. With the growing demand for advanced AI chips, Alibaba is well positioned to seize the opportunities presented by the digital age. Investing in infrastructure and continuous innovation are keys to the company’s success as it seeks to establish itself as a leader in the field of artificial intelligence.
