是说芯语·2025-10-27 07:42Core 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].