DeepSeek快速渗透，商业模式优势加速基础设施爆发

Investment Strategy - DeepSeek series models have achieved global leading performance and cost efficiency, potentially disrupting the high-end computing card market and creating opportunities for domestic computing chains[5] - The total training cost for DeepSeek-V3 is only $5.57 million, compared to $100 million for models like GPT-4o, indicating a significant cost advantage in AI model training[5] - The release of DeepSeek models is expected to lower the importance of high-end computing cards, benefiting domestic alternatives[5] Market Performance - DeepSeek's daily active users (DAU) reached 22.15 million, accounting for 41.6% of ChatGPT's DAU, indicating strong market acceptance[21] - Alphabet plans to increase its capital expenditure to $75 billion in 2025, up from $32.3 billion in 2023, driven by AI business growth[22] - The communication sector index rose by 2.69% this week, outperforming the Shanghai Composite Index which increased by 1.63%[26] Industry Developments - Major cloud providers, including Tencent Cloud and Alibaba Cloud, have integrated DeepSeek models into their services, enhancing AI capabilities across various applications[17] - The three major telecom operators in China have adopted DeepSeek models, indicating readiness for industry-wide AI integration[24] - The introduction of 5G millimeter-wave technology during the Asian Winter Games signifies advancements in commercial applications of this technology[25]