Chief Executive Eddie Wu outlined a sweeping strategy for AI models, chips, and data centers at the Apsara Conference in Hangzhou. The company intends to train new models with as many as 10 trillion parameters, utilizing the Zhenwu V900 AI chip scheduled for commercial release in Q1 2027. These chip clusters can scale to 500,000 units to support a global data center capacity target of 20 GW by 2032. Alibaba has committed more than $53 billion over three years to these infrastructure capabilities.

Qwen 4 is currently in training, with future generations 4.5 and 5 targeting the 10 trillion parameter mark. The company is exploring recursive self-improvement, a method allowing models to identify their own weaknesses and synthesize their own training data. Wu stated that machine thinking will eventually exceed all human thinking by 1,000 times, though current machine thought volume is less than 3% of the human total.