2025年08月25日 国泰君安期货商品研究晨报-贵金属及基本金属 观点与策略 | 黄金：JH会议鲍威尔放鸽 | 2 | | --- | --- | | 白银：冲顶前高 | 2 | | 铜：美元回落，价格上涨 | 4 | | 锌：区间震荡 | 6 | | 铅：库存减少，支撑价格 | 8 | | 锡：区间震荡 | 9 | | 铝：累库放缓 | 11 | | 氧化铝：横盘小涨 | 11 | | 铸造铝合金：跟随电解铝 | 11 | | 镍：窄幅区间震荡运行 | 13 | | 不锈钢：短线低位震荡 | 13 | 国 泰 君 安 期 货 研 究 所 请务必阅读正文之后的免责条款部分 1 期货研究 商 品 研 究 商 品 研 究 2025 年 8 月 25 日 黄金：JH 会议鲍威尔放鸽 白银：冲顶前高 | 王蓉 | 投资咨询从业资格号：Z0002529 | wangrong013179@gtjas.com | | --- | --- | --- | | 刘雨萱 | 投资咨询从业资格号：Z0020476 | liuyuxuan023982@gtjas.com | 【基本面跟踪】 贵金属基本面数据 | | | 昨日收盘价 ...

东方证券指出，DeepSeek正式发布V3.1版本，采用UE8M0 FP8参数精度，专为下一代国产芯片设计， 有望推动国产AI芯片在推理市场的规模化应用。随着国产大模型对国产芯片的适配和支持不断加强， 国产算力市占率有望持续提升，进一步增强投资者对国产AI生态的信心。此外，国产AI芯片在制造工 艺和技术设计方面持续进步，也为AI算力链的国产化替代提供了坚实基础。 中信建投证券认为，近期国产算力板块迎来密集催化，国产算力芯片迎来国产创新窗口期。 海外龙头方面，英伟达近日宣布推出NVIDIA Spectrum-XGS Ethernet技术。据介绍，这是一项面向分布 式数据中心的跨域互联技术，旨在将多个不同区域的独立数据中心连接到一起，打造成一个具备千亿级 计算能力的AI超级工厂。 Wind数据显示，8月25日，A股英伟达产业链指数盘中涨幅超2%，中科曙光涨停，奥比中光-UW、铂科 新材、中际旭创等个股涨幅居前。 场内ETF方面，截至2025年8月25日，港股通科技ETF(159262)一度涨超3%；成分股方面，ASMPT涨超 6%，快手-W、中兴通讯等跟涨。跟踪指数前十大权重股合计占比73.3%，权重股中芯国际阿 ...

国 泰 君 安 期 货 研 究 请务必阅读正文之后的免责条款部分 1 2025年08月25日 | 国泰君安期货商品研究晨报 | | --- | 观点与策略 | 黄金：JH会议鲍威尔放鸽 | 3 | | --- | --- | | 白银：冲顶前高 | 3 | | 铜：美元回落，价格上涨 | 5 | | 锌：区间震荡 | 7 | | 铅：库存减少，支撑价格 | 9 | | 锡：区间震荡 | 10 | | 铝：累库放缓 | 12 | | 氧化铝：横盘小涨 | 12 | | 铸造铝合金：跟随电解铝 | 12 | | 镍：窄幅区间震荡运行 | 14 | | 不锈钢：短线低位震荡 | 14 | | 碳酸锂：宽幅震荡 | 16 | | 工业硅：市场情绪提振 | 18 | | 多晶硅：区间震荡，以逢低做多为主 | 18 | | 铁矿石：短期估值仍有来自宏微观的支撑 | 20 | | 螺纹钢：宽幅震荡 | 22 | | 热轧卷板：宽幅震荡 | 22 | | 硅铁：宽幅震荡 | 24 | | 锰硅：宽幅震荡 | 24 | | 焦炭：宽幅震荡 | 26 | | 焦煤：宽幅震荡 | 26 | | 原木：震荡反复 | 28 | | 对 ...

英伟达宣布推出NVIDIA® Spectrum-XGS以太网。这项跨区域扩展（scale-across）技术可将多个分布式 数据中心组合成一个十亿瓦级AI超级工厂。 Spectrum-XGS以太网是NVIDIA Spectrum-X™ 以太网平台 新增的一项突破性产品，它通过引入跨区域扩展（scale-across）基础设施打破了上述限制。跨区域扩展 （scale-across）成为了继纵向扩展（scale-up）和横向扩展（scale-out）之后的 AI 计算"第三大支柱"， 能够将 Spectrum-X 以太网的极致性能和规模扩展至多个分布式数据中心，将它们组成具有十亿瓦级的 智能巨型 AI 超级工厂。 英伟达创始人兼首席执行官黄仁勋表示："AI 工业革命已经到来，而巨型 AI 工厂是这场变革的核心基础设施。继纵向扩展（scale-up）和横向扩展（scale-out）技术后，我们又通过 推出 NVIDIA Spectrum-XGS 以太网提供跨区域扩展（scale-across）技术，将不同城市、国家乃至大洲 的数据中心组合成庞大的十亿瓦级的 AI 超级工厂。"（NVIDIA英伟达网络微信公号） 风险 ...

Core Viewpoint - The fastest chip for running full-scale DeepSeek is a domestic GPU from Moore Threads, achieving a speed of 100 tokens/s, significantly faster than foreign GPUs at 50 tokens/s and domestic counterparts at 15 tokens/s [1][4]. Group 1: Moore Threads' Achievements - Moore Threads has developed an AI super factory that goes beyond just creating faster chips, focusing on a comprehensive transformation of the entire technology stack [6][10]. - The AI super factory is not a physical chip manufacturing facility but a systemic overhaul that includes innovations in chip architecture, cluster design, and software algorithms [9][10]. Group 2: Key Components of the AI Super Factory - The AI super factory's production efficiency is defined by five core elements: generality of accelerated computing, effective chip performance, node efficiency, cluster efficiency, and cluster stability [13]. - A full-function GPU serves as the foundation of the AI super factory, evolving from basic graphics acceleration to a versatile computing platform capable of handling various AI tasks [14][16]. Group 3: MUSA Architecture - The MUSA architecture acts as the "chief designer" of the super factory, allowing for scalable and configurable chip designs that optimize resource allocation [25][26]. - MUSA's innovative design enables global resource sharing, reducing bottlenecks and improving efficiency during multi-task operations [27][29]. Group 4: Full-Stack Software System - Moore Threads has created a full-stack software system that integrates deeply with the MUSA hardware architecture, enhancing developer experience and operational efficiency [35][36]. - The software stack includes optimized drivers, core operator libraries, and tools for performance analysis, significantly improving task handling and resource utilization [41][42]. Group 5: KUAE Computing Cluster - The KUAE computing cluster is a soft-hard integrated system that extends the performance advantages of individual GPUs to large-scale deployments, enabling efficient training of massive AI models [43][44]. - The cluster supports various parallel training strategies and provides end-to-end training optimization, ensuring high performance and stability [45][46]. Group 6: Zero-Interrupt Fault Tolerance Technology - Moore Threads has developed a unique zero-interrupt fault tolerance technology that allows for continuous operation of the AI super factory, minimizing downtime and recovery costs [47][49]. - This technology enhances the overall stability and reliability of the system, ensuring high effective training time and reducing the impact of potential failures [51][52]. Group 7: Future of AI and Computing Needs - The demand for computing power is expected to grow exponentially, driven by advancements in generative AI and the need for complex task execution [54][56]. - Moore Threads aims to provide a comprehensive solution that addresses the challenges of AI model training, emphasizing the importance of stability, reliability, and efficiency in future computing [58][61].

Core Viewpoint - The CEO of SAP, Christian Klein, argues that Europe does not need to rush into building numerous data centers to compete in the AI sector, contrasting with Nvidia CEO Jensen Huang's recent statements during his visit to Europe [1][2]. Group 1: AI Infrastructure and Investment - Klein questions the necessity of constructing five data centers equipped with top-tier chips, expressing skepticism about whether this is truly what Europe needs [2]. - He highlights that large language models, which require significant energy and computational power for training, are rapidly being commercialized, as demonstrated by the Chinese company DeepMind, which claims to have surpassed leading US AI developers at a low cost [2]. - The US has announced the "Stargate" initiative, planning to invest up to $500 billion, while the EU has committed to investing €20 billion (approximately $23 billion) to build five AI "super factories" dedicated to developing and training next-generation models [3]. Group 2: Strategic Focus for Europe - Klein suggests that European industries, such as automotive and chemicals, should focus on applying AI to enhance their operations rather than trying to catch up with the US in AI infrastructure [5]. - SAP has shifted its stance and is no longer seeking to be an operator or investor in AI super factory projects, but rather aims to provide technology and software support for potential future projects [4][5]. Group 3: Changes in Perspective - Klein's current viewpoint marks a shift from earlier this year when he referred to the Stargate project as an "excellent example" for Europe and expressed strong support for a European version of the initiative during the World Economic Forum in Davos [3].

Core Viewpoint - Germany plans to apply for at least two of the five AI super factories proposed by the European Union [1] Group 1 - The German Chancellor, Merz, emphasized the importance of AI development for the country's technological advancement [1] - The initiative aligns with the EU's broader strategy to enhance AI capabilities across member states [1] - The move is expected to strengthen Germany's position in the global AI landscape [1]

Core Insights - The article discusses NVIDIA's strategic shift from being a "GPU supplier" to a "global AI infrastructure provider," aiming to create "plug-and-play" AI factories [3][25] - Huang Renxun's keynote at Computex highlighted the need for NVIDIA to present stronger plans in light of a less optimistic outlook for the GeForce RTX 5060 [1][25] Product and Technology Highlights - NVIDIA introduced several key products, including the GeForce RTX 5060 GPU, Grace Blackwell GB300 supercomputing platform, and DGX Spark desktop AI supercomputer, aimed at enhancing performance for developers and gamers [4][6] - The RTX 5060 features 3,840 CUDA cores, a 17% reduction compared to the RTX 5060 Ti, and 8GB of memory, which has sparked discussions regarding its value proposition [4][6] - The DGX Spark, designed for developers and researchers, will soon be available and is equipped with the GB10 chip [6] - The RTX 5060 showcases DLSS neural rendering technology, where only 10% of the pixels are rendered, with the remaining 90% generated by AI, demonstrating significant advancements in rendering efficiency [7] Advanced Computing Solutions - The DGX Station, an upgraded version of Spark, is referred to as a "supercomputer" and is designed specifically for modern AI-native applications, eliminating the need for traditional IT software [9] - The Grace Blackwell supercomputing platform aims for scalability, with an upcoming upgrade to the GB300 model that will enhance memory by 1.5 times and network performance by two times [11] AI Infrastructure Vision - Huang Renxun's vision includes creating a "thinking machine" and a "giant GPU," positioning NVIDIA as an "AI super factory" that supports the entire AI ecosystem [14][25] - The NVLink Fusion product allows large-scale enterprises to create semi-custom computing solutions, breaking traditional data center bottlenecks and enabling optimized system designs for specific AI workloads [16] Robotics and Digital Agents - NVIDIA continues to focus on robotics, showcasing a Disney robotic dog powered by the Newton engine, which is set to be open-sourced in July [19] - The concept of "agentic AI" is introduced, representing AI versions of various professional roles, including digital marketing managers and software engineers [21] - The NVIDIA RTX PRO server series is part of the new enterprise-level AI Factory design, aimed at providing acceleration for AI, design, engineering, and business applications [23] Future Developments - NVIDIA plans to establish a new office in Taiwan, reinforcing its commitment to expanding its presence in the AI infrastructure space [23] - The company envisions AI as a foundational infrastructure akin to the internet and electricity, with the goal of becoming a leading AI super factory [25]