科技行业：DeepSeek之AI大模型解读：模型成本大幅下降，国产化全方位加速

Investment Rating - The report maintains an optimistic outlook on the AI computing chip industry, particularly for companies like TSMC, NVIDIA, and AMD, emphasizing TSMC as a top pick [8]. Core Insights - DeepSeek R1 model, launched on January 20, demonstrates capabilities comparable to OpenAI's models while significantly reducing costs, with API pricing set at 1 RMB per million input tokens (cache hit) and 4 RMB (cache miss) [3][4]. - The rapid growth in DeepSeek's app downloads, achieving a 375% week-on-week increase, indicates strong market interest and potential demand for AI applications [4]. - The report suggests that concerns over excessive AI computing power investments are overstated, as the AI large model industry is still in its early development stage, presenting growth opportunities for startups like DeepSeek [5][7]. Summary by Sections DeepSeek Model Development - DeepSeek R1 features 671 billion parameters and outperforms many existing models, with a pricing strategy that undercuts both domestic and international competitors [3][4]. - The model's performance has led to significant interest from major tech companies, including NVIDIA and Microsoft, which have integrated DeepSeek's technology into their AI platforms [4]. Market Dynamics - The report highlights that the demand for AI computing chips is expected to expand significantly due to declining costs and technological advancements, benefiting end-user applications [5][7]. - The ongoing geopolitical tensions may accelerate the growth of domestic AI enterprises in China, as they adapt to restrictions on advanced technologies from the U.S. [9][10]. Investment Opportunities - Companies in the smartphone and new energy vehicle sectors, such as Apple and Xiaomi, are expected to benefit from the lower costs associated with AI applications, enhancing their competitive edge [8]. - The report identifies potential beneficiaries in the AI driving application space, including NIO, XPeng, and BYD, as they leverage DeepSeek's advanced capabilities [9].