Investment Rating - The report maintains a "Buy" rating for the telecommunications industry and related companies, including specific recommendations for stocks such as China Telecom, Wolong Materials, and others [8][59]. Core Insights - Nvidia's acquisition of Groq for $20 billion highlights its commitment to the inference market and LPU products, indicating a potential shift in resource allocation towards these technologies [2][12][27]. - The LPU technology, designed for low-latency and high-efficiency real-time token generation, is expected to drive demand for copper cabling and liquid cooling solutions in the future [1][2][27]. - The report identifies a primary investment theme in the telecommunications sector focused on AI computing power, with recommended companies including Zhongji Xuchuang, New Yisheng, and others [3][59]. Summary by Sections Market Performance - The telecommunications index rose by 4.76% last week, outperforming the Shanghai Composite Index, which increased by 1.98%, and the Shenzhen Component Index, which rose by 2.80% [1][11]. Key Company Developments - Nvidia's integration plan for Groq's LPU technology aims to enhance its overall architecture, particularly in addressing low-latency decoding bottlenecks in AI applications [12][27]. - The LPU hardware is expected to be deployed in rack form, maintaining a focus on cost-effectiveness and energy efficiency, with copper connections being favored for short-distance interconnects [22][27]. Recommended Companies - The report recommends several companies for investment, including: - China Telecom (601728 CH) with a target price of 9.11 CNY - Wolong Materials (002130 CH) with a target price of 43.21 CNY - China Mobile (600941 CH) with a target price of 126.20 CNY - New Yisheng (300502 CH) with a target price of 476.71 CNY - StarNet Ruijie (002396 CH) with a target price of 39.16 CNY - Arista Networks (ANET US) with a target price of 166.30 USD - Zhongji Xuchuang (300308 CH) with a target price of 626.68 CNY - Ruijie Networks (301165 CH) with a target price of 102.51 CNY - Guanghuan New Network (300383 CH) with a target price of 18.87 CNY [8][59].

Core Viewpoint - ZTE Corporation has won a bid for the construction of a new AI inference resource pool for China Unicom in Henan, marking a significant advancement in telecom operators' layout in the AI computing power sector [1][6]. Group 1: Project Details - The project involves the procurement of 64 GPU servers and 26 RoCE switches, aimed at building an AI inference resource pool covering seven provinces in Central and Western China [2][5]. - The project budget is set at 74.26 million yuan (excluding tax), with ZTE winning the bid at 71.709 million yuan, achieving a cost saving of approximately 3.4% [2]. Group 2: Technical Aspects - The application of RoCE (RDMA over Converged Ethernet) technology is a highlight, enabling remote direct data access over Ethernet, significantly reducing data transmission latency within GPU clusters, which is crucial for efficient AI inference tasks [5]. - The project focuses on "inference resource pools," indicating that China Unicom aims to target the application of AI models in practical business scenarios rather than competing directly with major internet companies in large-scale model training [5]. Group 3: Market Positioning - This project is part of China Unicom's AI action plan and will support the demand for intelligent computing opportunities both within and outside the province [6]. - The launch of the Henan Unicom intelligent computing center project signifies an escalation in the competition among the three major telecom operators in the computing power infrastructure sector [6]. Group 4: Company Strategy and Future Orders - ZTE Corporation's successful bid was made in a consortium with Shenzhen ZTE Technology Service Company, showcasing its overall strength in the intelligent computing infrastructure field [8]. - In addition to the Henan project, ZTE has secured significant orders in the AI server procurement project with China Mobile, indicating a strong foothold in the AI computing market [10][11].