Venture(US:VEMLY) Mei Ri Jing Ji Xin Wen·2025-08-25 06:54Group 1 - DeepSeek released its latest large language model, DeepSeek-V3.1, on August 22, 2025, featuring a hybrid reasoning architecture and compatibility with next-generation domestic chip design parameters, driving technological upgrades in the AI industry [1] - Minsheng Securities highlighted that DeepSeek is leading the domestic AI computing power and top large model software-hardware collaboration, accelerating the development of the domestic AI industry [1] - The capabilities of domestic large models are continuously improving, with leading models like Doubao, DeepSeek, and Tongyi Qianwen reaching a global top level, providing a solid foundation for ecosystem construction [1] Group 2 - On the hardware side, several domestic chip manufacturers have supported FP8 precision and deeply adapted to large models, significantly enhancing training and inference efficiency through technological optimization [1] - In the context of major power technology competition, the development of domestic AI computing power is becoming an inevitable trend, with Huawei Ascend and other open-source ecosystems accelerating developer innovation [1] - The integration of leading large models with domestic chips lays the foundation for leapfrog development in the industry [1] Group 3 - In the Xinchuang sector, policies emphasize high-level technological self-reliance and continuous promotion of safe and reliable evaluations, with core manufacturers' products entering the market to accelerate industry growth [1] - The mid-year performance of listed companies generally improved, reflecting the release of demand, indicating that domestic AI computing power and the Xinchuang industry are likely to welcome a development turning point [1]