China pushes green compute and new domestic AI chips
China’s AI compute news centered on policy support for greener infrastructure, cloud vendors leaning on self-developed chips, and several new domestic chip and model launches.
• China’s cyberspace regulator said new computing power facilities should be co-located with renewable generation, and the action plan also promotes liquid cooling, waste heat recovery, and continued national green computing power facility construction.
• The same plan called for breakthroughs in key AI research, high-end AI chips, high-performance training clusters, and stronger foundation models such as large language, multimodal, and world models.
• Alibaba Cloud said it will rely more on self-developed chips to improve margins, and another report said its revenue rose 45% year on year while quarterly capex reached 67.6 billion yuan.
• Alibaba Cloud also added GLM-5.3 and DeepSeek-V4-Pro to the Qwen AI platform and opened the related API services.
• DeepSeek launched V4-Flash-Vision-Exp on its API platform, adding image input support for multimodal use cases.
• SenseTime open-sourced SenseNova U1.5 Lite, an 8B-parameter multimodal model with native 4K image output support.
• Qianhe Yibang said its four-layer 3D DRAM-stacked compute-in-memory chip has taped out and powered on, and the company recently completed a Series B round of more than 2 billion yuan.
• East Group plans to raise 682 million yuan for an integrated storage-compute fusion chip R&D project, while Zhongcheng Hualong released its HL200 inference chip and a supernode AI data center cluster plan.