China expands AI compute as chip and optical supply-chain activity picks up
China’s AI compute news centered on state-backed infrastructure targets, new large data-center plans, domestic-chip deployments, and supply-chain fundraising.
• The NDRC said China’s AI computing power at the end of June reached 2.8 times its level a year earlier, and that the country’s first fully domestically produced 100,000-card AI supercluster has entered service.
• The NDRC said computing power network construction is expected to generate 4 trillion yuan in new direct investment during the 15th Five-Year Plan period, with companies providing most of the funding.
• DeepSeek is reportedly planning a 1GW AI data center in Ulanqab, Inner Mongolia, with part of the computing power expected to come online by late 2027 or early 2028, and is also interested in leasing additional compute.
• Zhipu said it opened subscriptions for its GLM Coding Plan as infrastructure expands, while Yicai reported the company has completed a 1GW-level domestic AI computing power data center using domestic AI chips.
• Huawei open-sourced its 505-billion-parameter openPangu-2.0-Pro model, including weights, basic inference code and a technical report, describing it as trained on Ascend NPUs with Ascend-native training and inference technologies.
• Mingpu Opto-Magnetic plans to raise up to 1.283 billion yuan through a private placement for high-speed optical module, optical device, passive component and optical chip manufacturing projects.
• Montage Technology began trial production of a CXL 3.2 memory expansion controller chip supporting PCIe 6.x, CXL 3.2, transfer rates up to 64 GT/s and DDR5 memory up to 8000 MT/s.
• Zhongcheng Hualong won China Telecom Tianyi Cloud and China Mobile server-related projects worth a combined 130 million yuan, ranking first overall in both bids.