Alibaba ships Qwen update as Chinese compute vendors adapt MiniMax H3
China’s AI compute news centered on new foundation-model releases, domestic accelerator support, data-center investment plans and memory-sector financing.
• Alibaba released Qwen3.8 on August 3 with 2.4 trillion parameters, opened its API on the Qwen AI platform, integrated it into QwenWork, and said Qwen3.8-Max weights are expected to be open-sourced next week.
• The National Supercomputing Internet launched API access for Alibaba’s Qwen3.8-Max and API access plus download services for DeepSeek-V4-Flash, whose official API recently entered public beta.
• MiniMax open-sourced its H3 multimodal generative model, and Moore Threads, MetaX and Biren announced day-0 adaptation or inference validation on their respective AI compute platforms.
• Xingyun Technology said its 3- to 5-year framework orders for computing power and storage exceeded 15.404 billion yuan as of the end of July 2026, with approved comprehensive credit lines totaling 18.024 billion yuan.
• Sunshine Co. said its Shenzhen Sunshine Digital Technology subsidiary plans to invest up to 980 million yuan in the Sunshine AI Data Center project in Xingning, Meizhou, Guangdong, with operations expected to begin in October 2027.
• CXMT raised its registered capital by about 31% to roughly 31.39 billion yuan, while TrendForce forecast PC DRAM contract prices would rise 15% to 20% quarter over quarter in the third quarter because of supply shortages.
• Fangqing Technology completed an A1 financing round at a valuation above 10 billion yuan, with proceeds mainly for self-developed chips and systems, mass production, software ecosystem development and hiring.
• Liangyin Technology raised tens of millions of yuan in angel funding for silicon photonics transmission chips, Optical IO and co-packaged optics, while Westlake Xinchen completed a several-hundred-million-yuan B+ round for diffusion language model R&D, multimodal products and overseas expansion.