【兴证计算机】DeepSeek（深度）：加速迭代的开源大模型引领者

Group 1 - DeepSeek aims to reshape the AI industry landscape by leading the open-source model ecosystem, breaking the monopoly of foreign models with its DeepSeek-R1 inference model launched on January 20, 2025, which features lower training costs and superior model capabilities [1][6] - The organization has a flat structure with a young team of fewer than 140 members, primarily composed of top talent from domestic universities [1] - DeepSeek adheres to a long-termism philosophy, avoiding short-term commercial interests, and is projected to reach a valuation of 1.05 trillion yuan by September 30, 2025 [1] Group 2 - DeepSeek is focused on optimizing model training and deployment costs, with the upcoming DeepSeek-V3 model set to launch in December 2024, featuring a training cost of only 5.57 million USD and performance comparable to GPT-4 [2] - The DeepSeek-R1 model, released in January 2025, offers leading inference capabilities and facilitates low-cost private deployment, significantly advancing AI applications in finance, healthcare, and government sectors [2] Group 3 - The anticipated release of the V4 model in February is expected to enhance code generation and processing capabilities, surpassing existing models like Claude and GPT [3] - The commercial potential of DeepSeek's ecosystem is seen as a core barrier to the commercialization of large models, with the company positioned to become a foundational infrastructure for AI applications across various industries [3] Group 4 - Investment opportunities are suggested in AI applications and computing power sectors, with specific companies listed for potential investment, including 卓易信息, 汉得信息, and 海光信息 among others [4]