AI动态跟踪系列（四）：DeepSeek引发广泛关注，大模型应用落地将加速

Investment Rating - The industry investment rating is "Outperform the Market" [52] Core Insights - The release of DeepSeek-V3 and DeepSeek-R1 indicates that domestic large model capabilities are now comparable to leading overseas models, with lower costs and high performance expected to accelerate the deployment of large models in both B-end and C-end application scenarios [5][48][30] - DeepSeek's open-source, low-cost, and high-performance models are anticipated to significantly reduce the costs associated with obtaining, deploying, and applying large models, thereby enhancing their adoption across various industries [30][5] - The global attention garnered by DeepSeek's models is expected to impact the competitive landscape of the global large model industry, challenging the dominance of overseas leading model manufacturers [30][5] Summary by Sections Section 1: DeepSeek Model Releases - DeepSeek-V3, released on December 26, 2024, features a self-developed MoE model with 671 billion parameters and has shown superior performance compared to other open-source models [6][3] - The training cost for DeepSeek-V3 was approximately $557,600, significantly lower than comparable models like Llama 3 [8][9] - DeepSeek-R1, released on January 20, 2025, aligns in performance with OpenAI's o1, showcasing the competitive edge of domestic models [12][17] Section 2: Global Attention and Adoption - DeepSeek's AI assistant app became the most downloaded mobile application in 140 markets shortly after its launch, achieving 16 million downloads in just 18 days [21][22] - Major tech companies, including NVIDIA and Microsoft, have integrated DeepSeek models into their platforms, indicating strong industry acceptance [25][27] Section 3: Market Impact and Future Demand - The introduction of DeepSeek models is expected to maintain the upward trend in overall AI computing demand, particularly in inference and edge computing [31][36] - The low-cost and open-source nature of DeepSeek models is projected to lower the barriers for AI application across various sectors, enhancing the potential for rapid growth in inference computing needs [36][31] Section 4: Investment Recommendations - The report recommends several companies for investment based on their involvement in AI infrastructure and applications, including Inspur Information, Sugon, and iFLYTEK, among others [48]