China pushes computing-power buildout as major model and cloud updates land
China's AI compute news centered on state-led infrastructure planning, rising power needs, cloud GPU capacity, and new large-model deployments from DeepSeek and Alibaba.
• SASAC chairman Cheng Fubo visited China Telecom and called for stronger new infrastructure, coordinated development of the national integrated computing power network, and breakthroughs in cloud computing, networks, AI and quantum technologies.
• A CCTV Finance-cited report said a 10,000-GPU cluster consumes at least 200,000 kWh of electricity per day, while China's computing-power electricity use is estimated to reach 800 billion kWh by 2030.
• Tencent said GPU resources should become more ample from late this year to early next year, supporting both internal model development and external cloud leasing, while large AI-related infrastructure prepayments contributed to negative second-quarter free cash flow.
• DeepSeek updated its API documentation with DeepSeek-V4-Pro-0813, keeping the invocation method unchanged and adding support for non-thinking and thinking modes, Responses API and Codex integration.
• Alibaba opened weights for Qwen3.8-2.4T-A95B, a Qwen-Max-class MoE model with 2.4 trillion total parameters, 95 billion active parameters per token and a native 262,144-token context window.
• Alibaba Cloud launched the Lingjun Zhenwu M890 supernode instance in Ulanqab, saying it can support inference for MoE models with up to 10 trillion parameters and has run a model with more than 2 trillion parameters.
• Cambricon said demand for intelligent computing continues to rise and that it will focus on AI chip design, market expansion and real-world deployment after its first-half earnings briefing.
• Sinosun Technology plans to acquire control of Bainn Technology, a provider of testing, repair, maintenance and lifecycle technical services for electronic equipment and server components, as it enters the server infrastructure services sector.