At the 2026 Yunqi Conference in Hangzhou, Alibaba has laid out its most ambitious vision for artificial intelligence. CEO Wu Yongming announced that the company is preparing to train a new generation of models with parameters ranging from 5 trillion to 10 trillion, which is the boldest AI development plan in Alibaba's history and signals its transformation from a cloud computing service provider into a full-stack AI platform.

According to the disclosure, this newly developed model will be an important follow-up product of the Qwen series. The Qwen team is currently continuously pushing forward architectural innovation and data optimization, aiming to enable future models to handle more complex and longer-term tasks, and move towards artificial superintelligence (ASI). Currently, Alibaba's strongest flagship model, Qwen 3.8 Max, has about 2.4 trillion parameters, while the planned new model will have two to four times that size, placing it among the largest-scale AI systems in the world.

More intriguingly, Wu Yongming revealed progress on recursive self-improvement (RSI). In this mode, the model can actively identify its own capability boundaries, design experiments, generate data, and continuously optimize itself, forming a closed loop of self-evolution. Alibaba believes that this capability will become an essential foundation for more advanced AI systems—meaning the model is not only being trained but also learning to act as its own coach.

On the chip side, Alibaba has also made significant investments. Alibaba simultaneously launched the new AI chip Zhenwu V900, developed by the semiconductor team of TSMC. Wu Yongming called it the most powerful AI chip in China currently, with three times the computing power of its predecessor M890. A single AI cluster built based on the Zhenwu V900 can support up to 500,000 accelerator cards working in coordination, focusing on the training and inference of cutting-edge large models; this chip is expected to mass-produce and officially commercialize in the first quarter of 2027. As a comparison, the existing M890 AI super node can already support inference for large models with over 2 trillion parameters, and Wu Yongming frankly stated that very few companies globally can achieve this capability.

The infrastructure expansion plan is also grand. Alibaba plans to increase the total capacity of its global data centers to over 20 gigawatts by 2032, further expanding AI computing power supply. Enterprise customers' demand for AI services has been seen by Alibaba as a key engine for revenue growth, but the company does not hide the fact that the global supply chain is constrained by insufficient equipment, power, and related infrastructure in data centers, limiting the speed of expansion. Wu Yongming assesses that in the medium to long term, industry demand growth still far exceeds supply, and Alibaba is accelerating the deployment of AI super nodes, starting from this quarter, making large-scale commercial operations.

From a more macro perspective, Wu Yongming compared the current turning point to the level of the Industrial Revolution, calling it the beginning of the era of machine intelligence. He did some calculations: the total amount of thinking generated by machines is still less than 3% of all human thinking activities, but in the future, it could grow to more than 1000 times the total human thinking. The products and applications that truly define this era may not yet exist. He compared the current popular AI programming applications to the early light bulbs of the electrical era—just the beginning, not the end. As model capabilities continue to strengthen, artificial intelligence will take over complex decision-making and long-term task execution in more fields.

Amid the ongoing tightening of high-performance chip export restrictions in the United States, Chinese tech companies are accelerating the construction of an independent AI industrial chain. Alibaba's simultaneous advancement of super-large models, self-developed chips, and data centers is seen as a critical step in building a complete AI ecosystem, indicating that the global competition surrounding super-large models, computing infrastructure, and next-generation AI chips has entered a new stage.