China moves on zero-carbon compute as AI firms and model policy draw attention
China’s AI compute news centered on new energy rules for computing facilities, server procurement, foundation-model updates, and debate over Chinese open-source models.
• China’s Ministry of Industry and Information Technology launched a national zero-carbon factory program that includes zero-carbon computing power facilities, with non-fossil energy required to account for at least 95% of consumption by 2030.
• Entive Smart said its wholly owned subsidiary Gansu Yisuan plans to buy up to 2 billion yuan of servers and supporting equipment from multiple suppliers to provide computing power services.
• A DeepSeek insider said claims that leaked investor-meeting minutes affected the company’s financing were “not credible” and that most outside information is false or speculation, while DeepSeek has not issued an official response.
• Moonshot AI reportedly held a Beijing celebration after Kimi K3 drew outside attention, with leaked materials describing K3 as a “capacity expansion and upgrade” and calling for K4 to be pushed further.
• A weekly AI roundup said Alibaba Tongyi Qianwen released Qwen-Image-3.0, Huawei showed its full computing module lineup for the first time, Tencent Hunyuan launched the Hyra-1.0 agent, and CAS-inspired Brain Computing released an integrated computing-and-electricity platform.
• Another weekly applications roundup said ByteDance released the Seed Audio 1.0 audio model, Tencent merged its Hunyuan teams into a foundation model department, and Ant released Agentar 2.0.
• The New York Times reported that OpenAI and Anthropic have lobbied the U.S. to restrict Chinese open-source models, while NVIDIA’s Jensen Huang and Elon Musk have publicly opposed such restrictions.
• NVIDIA CEO Jensen Huang told Bloomberg that China is bound to produce outstanding AI technology and said China may be training more AI talent each year than all other countries combined.