创业板人工智能ETF国泰（159388）跟踪的是创业板人工智能指数（970070），单日涨跌幅达 20%，该指数从创业板市场中选取涉及人工智能技术开发与应用的企业证券作为指数样本，覆盖软硬件 研发、智能系统集成及相关服务等领域，以反映创业板市场内AI相关上市公司证券的整体表现。行业 配置上，该指数侧重信息技术和智能制造方向。 【免责声明】本文仅代表作者本人观点，与和讯网无关。和讯网站对文中陈述、观点判断保持中立，不对所包含内容 的准确性、可靠性或完整性提供任何明示或暗示的保证。请读者仅作参考，并请自行承担全部责任。邮箱： news_center@staff.hexun.com 每日经济新闻 兴业证券指出，AI互连推动通信行业高景气，Credo公司FY26 Q2收入同比+272%、环比+20%，主 要由AI数据中心互连快速放量驱动，毛利率维持67.7%的行业领先水平。DCI需求爆发叠加成本上涨， 驱动相干光模块陶瓷管壳开启涨价，成为又一紧缺环节；伴随800G、1.6T光模块放量，数据中心光模 块陶瓷基板业务有望爆发式增长。AI集群规模扩大带动核心客户BOM份额持续提升，客户粘性显著增 强。 （责任编辑：张晓波 ...

Core Insights - The article discusses the exponential growth in demand for high-speed data transmission in data centers driven by advancements in AI, cloud computing, and autonomous driving, necessitating a fundamental iteration of interconnect architectures [1][3]. Group 1: Industry Trends and Requirements - AI/ML data centers represent a complex "network of networks," requiring low power, low latency, and high density in data transmission, which presents unprecedented interconnect bandwidth challenges [3][4]. - By 2025, training models with hundreds of billions to trillions of parameters will require exabyte-level data processing capabilities, with current interconnect solutions facing limitations [3][4]. - The transition from 400G to 448G interconnect technology is critical to meet the next generation's bandwidth, latency, and scalability demands [3][4]. Group 2: Technological Evolution - The evolution of data transmission rates has been a continuous process over the past two decades, with OIF leading the development of standards from 0G to 448G [6][7]. - The article outlines the historical progression of interconnect speeds, highlighting key milestones such as 6Gbps in 2004, 56Gbps in 2017, and the upcoming 112Gbps in 2024 [6][7][8]. - Innovations in modulation technology, such as PAM (Pulse Amplitude Modulation), have significantly reduced bandwidth requirements while increasing data rates [8][11]. Group 3: Industry Collaboration and Standardization - Global standard organizations and industry players are actively collaborating on the research and standardization of 448Gbps/lane interconnect technology to push the limits of current computing infrastructure [4][14]. - The OIF has initiated the CEI-G framework project to establish a solid foundation for 448G technology, with a formal standard expected by 2026 [12][39]. - Industry events, such as the ODCC and OCP Global Summit, are facilitating discussions among key players like Tencent, Huawei, and Google to explore the technical pathways for 448G [15][19][23]. Group 4: Technical Challenges and Solutions - The article highlights the dual pressures of power efficiency and density in the 448Gbps era, with OIF identifying pJ/bit as a critical metric for power consumption [42][43]. - Solutions for thermal management are evolving, with companies like Ciena and Huawei developing liquid cooling technologies to address high thermal densities [42][43]. - The complexity of system optimization is underscored by the need for consensus on modulation schemes and the balance between electrical and optical transmission [43][46]. Group 5: Future Outlook - The successful commercialization of 448Gbps technology is seen as essential for overcoming data transmission bottlenecks and reducing communication overhead in AI training [39][40]. - The article concludes that the collaborative efforts across the industry will be crucial in achieving scalable commercialization of 448Gbps technology within the next 2-3 years, supporting the next wave of AI and digital economy growth [47][48].