Group 1 - The core viewpoint indicates that the demand for AI computing power interconnection is rapidly increasing, with a notable growth in the configuration ratio of Sci-Tech Innovation Board stocks in actively managed equity funds reaching a historical high in Q2 [2][3] - The Sci-Tech AI ETF (588790) has shown significant performance, with a recent increase of 0.68% and a weekly growth of 0.85%, indicating strong investor interest [2][3] - The total scale of Sci-Tech Board thematic funds has surpassed 300 billion, suggesting a growing market for hard technology investments [2] Group 2 - The Sci-Tech AI ETF has experienced a notable increase in scale, growing by 28.22 million in the past week, ranking first among comparable funds [3] - The ETF has also seen a significant increase in shares, with a growth of 45 million shares in the same period, indicating strong demand [3] - Despite a recent net outflow of 10.64 million, the ETF has attracted a total of 53.09 million in net inflows over the last five trading days [3] Group 3 - Leveraged funds have been actively buying into the Sci-Tech AI ETF, with a maximum single-day net purchase of 148 million, leading to a current financing balance of 476 million [4] - The ETF has achieved a net value increase of 9.21% over the past six months, ranking first among comparable funds [4] - Historical performance shows a maximum monthly return of 15.59% since inception, with a 100% probability of profit over a six-month holding period [4] Group 4 - The management fee for the Sci-Tech AI ETF is 0.50%, and the custody fee is 0.10%, which are relatively low compared to comparable funds [5] - The tracking error for the ETF over the past month is 0.014%, indicating high tracking precision [5] - The underlying index, the Sci-Tech Innovation Board AI Index, consists of 30 large-cap stocks that provide foundational resources and technology for the AI industry, with the top ten stocks accounting for 68.03% of the index [5]

Investment Rating - The industry investment rating is "Overweight" as the report anticipates that the interconnection bandwidth demand will experience rapid growth due to the "multiplier effect" [4][90]. Core Insights - The report identifies four major trends in the AI computing interconnection sector: 1. Computing hardware demand is shifting from being anchored to cloud vendors' CapEx to being anchored to the number of tokens processed. 2. The landscape of computing chips is evolving from reliance on GPUs to a coexistence of GPUs and ASICs. 3. Technological iterations are transitioning from product-level upgrades (like optical modules, PCBs, switches, and network cards) to overall upgrades at the network architecture level. 4. The focus of network interconnection upgrades is expanding from Scale Out to Scale Up, enhancing both types of networks and increasing single-chip bandwidth [5][81][85]. Summary by Sections Section 1: Market Trends - The demand for computing power is increasingly linked to the number of tokens processed rather than just cloud vendor capital expenditures [5][81]. - The growth of AI models has led to a significant increase in the demand for high-bandwidth, low-latency interconnections, particularly in the context of large-scale AI training [21][25]. Section 2: Product Innovations - New products like Sora and GB200 are driving demand for multimodal inference, with Sora requiring significantly more inference computing power compared to previous models [27][35]. - The report highlights that the performance of new interconnection technologies, such as GB200, is substantially superior to older models, enhancing overall computational efficiency [75][76]. Section 3: Capital Expenditure Trends - Capital expenditures from major cloud service providers (CSPs) are expected to continue increasing, with a notable shift towards AI computing capabilities [35][37]. - The report estimates that the ratio of NVIDIA's data center revenue from cloud vendors to the capital expenditures of five major cloud vendors has increased from 5% in Q1 2023 to approximately 30% by the second half of 2024 [37][38]. Section 4: Investment Opportunities - The report suggests that various interconnection methods, including optical, copper, and PCB connections, are likely to benefit from the anticipated growth in interconnection demand. Key companies identified include: - Optical interconnection: Zhongji Xuchuang, Xinyi Sheng, Tianfu Communication, Yuanjie Technology, Changguang Huaxin, Changxin Bochuang, Shijia Photon, Zhizhi Technology, Taicheng Light. - Copper interconnection: Zhaolong Interconnection, Zhongji Xuchuang, Hongteng Precision. - PCB: Shenghong Technology, Jingwang Electronics, Hudian Co., Shengyi Technology [4][90].