Investment Rating - The industry rating is "Outperform" (maintained rating) [8] Core Insights - The semiconductor industry outlook for Q3 is optimistic, with a focus on domestic computing power and self-controllable directions. The release of DeepSeek V3.1 model is expected to catalyze industry turning points by optimizing for next-generation domestic chips [2][16] - Nvidia's H20 chip sales plans may change, highlighting the long-term importance of establishing a domestic computing chip supply chain. Domestic model development companies are expected to increase procurement and usage of domestic chips [3][17] - The global AI smart glasses market is projected to see significant growth, with a 110% year-on-year increase in shipments in H1 2025. Major companies are expected to launch new AI smart glasses, driving demand [4][18] Summary by Sections Q3 Semiconductor Outlook - The release of DeepSeek V3.1 is designed for next-generation domestic chips, enhancing the synergy between domestic computing power and models, which is expected to accelerate the self-controllable process in the industry [2][16] - Nvidia has paused production of H20 chips for the Chinese market, indicating a shift towards developing a new AI chip based on the Blackwell architecture, which is expected to outperform H20 [3][17] July Semiconductor Industry Data - In July, the semiconductor supply chain showed stable growth, with rising orders in wafer foundry and packaging testing sectors. The overall delivery times are expected to increase, with prices continuing to rise [5][19] - The semiconductor market is entering a recovery phase, with global sales projected to reach approximately $626.87 billion in 2024, reflecting a 19% year-on-year growth [38][39] Key Recommendations - Focus on domestic chip design companies such as Cambricon, Haiguang Information, and Chipone Technology, as well as foundry and packaging companies like SMIC and Hua Hong Semiconductor [3][17][7] - In the AI SoC and storage segments, companies like Hengxuan Technology and Jiangbolong are recommended due to their potential in adapting to computing power needs [4][18][37]