Core Viewpoint - UBS reduced its stake in ZTE Corporation (000063)(00763) by selling 3,718,096 shares at a price of HKD 29.0796 per share, totaling approximately HKD 108 million, resulting in a new holding of about 60,298,700 shares, representing 7.98% of the company [1] Summary by Category - **Share Reduction** - UBS sold 3,718,096 shares of ZTE Corporation [1] - The sale price was HKD 29.0796 per share [1] - Total proceeds from the sale were approximately HKD 108 million [1] - **Post-Transaction Holdings** - After the reduction, UBS's remaining shares in ZTE Corporation are approximately 60,298,700 [1] - The new ownership percentage is 7.98% [1]

智通财经APP获悉，香港联交所最新资料显示，8月20日，瑞银减持中兴通讯(00763)371.8096万股，每 股作价29.0796港元，总金额约为1.08亿港元。减持后最新持股数目约为6029.87万股，最新持股比例为 7.98%。 ...

Group 1 - On August 26, 2025, the top three companies by trading volume in the Southbound Stock Connect were Yingfu Fund (02800) with a trading volume of 5.01 billion, SMIC (00981) with 3.86 billion, and Alibaba-W (09988) with 3.29 billion [1] - In the Southbound Stock Connect for Shenzhen-Hong Kong, the top three companies were Alibaba-W (09988) with a trading volume of 2.19 billion, SMIC (00981) with 1.95 billion, and Tencent Holdings (00700) with 1.81 billion [1] Group 2 - The top active companies in the Southbound Stock Connect (Hong Kong) included Yingfu Fund (02800) with a net buy of 4.997 billion, Alibaba-W (09988) with a net buy of 448 million, and Tencent Holdings (00700) with a net buy of 911 million [2] - In the Southbound Stock Connect (Shenzhen), Alibaba-W (09988) had a net buy of 897 million, while SMIC (00981) and Tencent Holdings (00700) had net sells of 275 million and 240 million respectively [2]

格隆汇8月26日｜香港交易所信息显示，摩根大通在中兴通讯的持股比例于08月21日从6.27%升至 6.98%。 ...

据香港交易所披露，花旗集团对中兴通讯股份有限公司-H股的多头持仓比例于2025年8月20日从5.05% 增至5.3%。摩根大通(JPMorgan)对中兴通讯股份有限公司-H股的多头持仓比例于2025年8月21日从6.27% 增至6.98%。瑞银集团(UBS)对中兴通讯股份有限公司-H股的多头持仓比例于2025年8月20日从8.47%降 至7.98%。 ...

Core Viewpoint - The continuous decline in the number of shareholders for many companies indicates a trend of increasing concentration of shares, with 582 companies reporting shareholder numbers as of August 20, and 92 companies experiencing a decline for more than three consecutive periods [1][2]. Shareholder Trends - 92 companies have seen their shareholder numbers decrease for more than three consecutive periods, with the most significant decline being 13 periods for Unisoc Microelectronics, which has seen a cumulative decrease of 27.88% in shareholder numbers [1]. - ZTE Corporation has experienced a decline for 12 periods, with a current shareholder count of 455,100 and a cumulative decrease of 16.17% [1]. - Other companies with notable declines include Tianhe Co., Huaping Co., and ST Huawen, among others [1]. Market Performance - Among the companies with declining shareholder numbers, 65 have seen their stock prices rise, while 25 have experienced declines. Notable gainers include Yidong Electronics, ST Jinglun, and Chuangyitong, with increases of 62.20%, 49.44%, and 39.79%, respectively [2]. - 27 companies, accounting for 29.35%, outperformed the Shanghai Composite Index during this period, with Yidong Electronics, ST Jinglun, and Chuangyitong achieving relative returns of 51.98%, 35.01%, and 27.48% compared to the index [2]. Industry Distribution - The industries with the highest concentration of companies experiencing declining shareholder numbers include basic chemicals, machinery equipment, and electronics, with 13, 10, and 8 companies, respectively [2]. - The distribution of these companies shows that 60 are listed on the main board, 30 on the ChiNext board, and 2 on the Sci-Tech Innovation board [2]. Institutional Activity - In the past month, 4 companies with declining shareholder numbers have been investigated by institutions, with Jieya Co. being the most frequently researched, receiving 2 inquiries [2]. - The companies with the highest number of institutional inquiries include Unisoc Microelectronics, Fujilai, and Jieya Co., with 124, 24, and 17 institutions participating in their research, respectively [2]. Financial Performance - 39 companies have reported their semi-annual results, with Fujilai showing the highest year-on-year net profit growth of 12,430.96% [3]. - Among the companies that have released their interim performance forecasts, Yalian Machinery and Donghua Technology reported net profit growth rates of 20.57% and 14.64%, respectively [3]. - 18 companies have issued performance forecasts, with 4 expecting an increase and 3 predicting profitability [3].

| 指数 | 最新 | 涨跌幅 | | --- | --- | --- | | 相生指数 | 25524.92 | -1.18% | | 国企指数 | 9148.66 | -1.07% | | 红筹指数 | 4343.85 | -1.02% | 免责声明：本文内容与数据仅供参考，不构成投资建议，使用前请核实。据此操作，风险自担。 每经AI快讯：北京时间8月26日16:00，恒生指数收盘下跌304.99点，跌幅为1.18%，报收25524.92点；国 企指数收盘下跌99.34点，跌幅为1.07%，报收9148.66点；红筹指数收盘下跌44.65点，跌幅为1.02%，报 收4343.85点。南向资金当日净流入165.73亿港元。东方甄选跌超11%，蔚来、君实生物跌超6%，中兴 通讯跌近6%，碧桂园跌超5%；中国黄金国际涨超10%，灵宝黄金涨超9%。 （记者 胡玲） ...

Core Insights - The article discusses the challenges of expanding Key-Value Cache (KV Cache) storage in large language models (LLMs), highlighting the conflict between reasoning efficiency and memory cost [1] - It emphasizes the need for innovative solutions to enhance KV Cache storage without compromising performance [1] Industry Exploration - Nvidia's Dynamo project implements a multi-level caching algorithm for storage systems, but faces complexities in data migration and latency issues [2] - Microsoft's LMCahce system is compatible with inference frameworks but has limitations in distributed storage support and space capacity [3] - Alibaba proposed a remote storage solution extending KV Cache to Tair database, which offers easy scalability but struggles with low-latency requirements for LLM inference [3] Emerging Technologies - CXL (Compute Express Link) is presented as a promising high-speed interconnect technology that could alleviate memory bottlenecks in AI and high-performance computing [5] - Research on using CXL to accelerate LLM inference is still limited, indicating a significant opportunity for exploration [5] Mariana Exploration - ZTE Corporation and East China Normal University introduced a distributed shared KV storage technology named Mariana, which is designed for high-performance distributed KV indexing [6] - Mariana's architecture is tailored for GPU and KV Cache storage, achieving 1.7 times higher throughput and 23% lower tail latency compared to existing solutions [6] Key Innovations of Mariana - The Multi-Slot lock-based Concurrency Scheme (MSCS) allows fine-grained concurrency control at the entry level, significantly reducing contention and improving throughput [8] - Tailored Leaf Node (TLN) design optimizes data layout for faster access, enhancing read speeds by allowing simultaneous loading of key arrays into SIMD registers [10] - An adaptive caching strategy using Count-Min Sketch algorithm identifies and caches hot data efficiently, improving read performance [11] Application Validation - Mariana's architecture supports large-capacity storage by distributing data across remote memory pools, theoretically allowing unlimited storage space [13] - Experimental results indicate that Mariana significantly improves read/write throughput and latency performance in KV Cache scenarios [14] Future Prospects - Mariana's design is compatible with future CXL hardware, allowing seamless migration and utilization of CXL's advantages [18] - The advancements in Mariana and CXL technology could lead to efficient operation of large models on standard hardware, democratizing AI capabilities across various applications [18]

Investment Strategy - Concerns over Nvidia's security may accelerate the replacement of domestic chips, creating a vacuum in the supply of computing chips in China. Recent advancements in domestic computing cards are expected to drive rapid development in the related industry chain. Nvidia's H20 computing chip has been reported to have serious security issues, which could negatively impact its future orders in the Chinese market. In 2024, over 50% of the data center accelerator cards in China are expected to be inference cards, with domestic computing cards projected to account for over 40% by the first half of 2025. The domestic accelerated computing server market is expected to grow by 97.3% year-on-year in 2024 and by 52.9% in 2025 [10][12][13] - The release of DeepSeek-V3.1 is expected to enhance the usability of domestic computing cards in large model applications. The V3.1 version supports UE8M0 FP8 precision, which significantly reduces storage and improves computational throughput compared to traditional FP16 and FP32 formats. This advancement is anticipated to lead to a qualitative leap in the application of domestic computing cards [11][14] - The demand for computing infrastructure from operators and cloud vendors is clear, with capital expenditures expected to increase in the second half of the year. The three major operators reported good growth in computing-related business revenues, with China Telecom's cloud service revenue at 57.3 billion yuan, China Unicom's at 37.6 billion yuan, and China Mobile's at 56.1 billion yuan. Capital expenditures for the first half of 2025 are projected to be 494 billion yuan for China Telecom, 348 billion yuan for China Unicom, and 928 billion yuan for China Mobile, accounting for 59.1%, 63.3%, and 61.4% of their annual expenditures respectively [12][13][18] Industry News - The successful launch of the low-orbit satellite internet 09 group marks a significant step in China's satellite internet development, with the country entering a high-density launch phase. This mission was the 590th flight of the Long March series of rockets, and it is expected to enhance the satellite internet network [15][16] - The budget for the 2025 rocket launch service procurement project by Shanghai Yuanxin Satellite Technology Co., Ltd. is approximately 1.336 billion yuan, with a focus on launching 94 satellites across seven missions. This indicates a growing emphasis on satellite internet and commercial space development in China [16][17] - Capital expenditures by the three major operators in the first half of 2025 were the lowest in recent years, totaling 112.8 billion yuan, a decrease of 16.5% year-on-year. However, there is an expectation of increased investment in AI and computing infrastructure in the second half of the year, which could significantly boost the domestic computing chain [18][19] Weekly Review and Focus - The communication sector saw a significant increase of 10.47% this week, outperforming major indices. The IDC and optical module sectors continued to lead gains, with increases of 14.59% and 14.41% respectively. The overall performance of the communication sector suggests strong investment opportunities in domestic computing chains and satellite internet [19][20]

8月25日恒生指数上涨1.94%，南向资金全天合计成交金额为2024.76亿港元，其中，买入成交1005.50亿 港元，卖出成交1019.26亿港元，合计净卖出金额13.76亿港元。具体来看，港股通（深）累计成交金额 794.06亿港元，买入成交402.33亿港元，卖出成交391.73亿港元，合计净买入金额10.60亿港元；港股通 （沪）累计成交金额1230.70亿港元，买入成交603.17亿港元，卖出成交627.53亿港元，合计净卖出金额 24.36亿港元。 成交活跃股方面，今日上榜个股中，南向资金成交金额最多的是中芯国际，合计成交额116.29亿港元， 阿里巴巴-W、腾讯控股成交额紧随其后，分别成交111.44亿港元、74.45亿港元。以净买卖金额统计， 净买入的个股共有7只，阿里巴巴-W净买入额为5.87亿港元，净买入金额居首，该股收盘股价上涨 5.51%，快手-W净买入额为4.72亿港元，中兴通讯净买入额为4.08亿港元。净卖出金额最多的是盈富基 金，净卖出23.52亿港元，该股收盘股价上涨2.01%，小米集团-W、中芯国际遭净卖出15.25亿港元、 8.11亿港元。 今日上榜个股中，阿里巴巴-W、中 ...