中信建投：继续看好国产算力产业链

Core Viewpoint - The release of DeepSeek-V3.1, which employs a hybrid reasoning architecture supporting both "thinking" and "non-thinking," is expected to enhance performance in programming and search applications, indicating strong competitiveness of domestic large models in the market [1] Group 1: Product Development - DeepSeek-V3.1 utilizes UE8M0FP8Scale parameter precision, specifically designed for the upcoming generation of domestic chips [1] - The new large model has shown significant improvements in performance after training, particularly in agent applications [1] Group 2: Market Outlook - According to CITIC Securities, the competitive strength of domestic large models like DeepSeek is likely to drive growth in computing power demand [1] - The firm maintains a positive outlook on the domestic computing power industry chain [1]