2024-2025年，英伟达陆续推出GH200 NVL72、GB200/ GB300 NVL72等成熟超节点解决方案。根据大 摩预测，2025年英伟达 GB200/300 NVL72出货量约2800台。展望2026-2027年，英伟达计划推出 Vera Rubin NVL144和Rubin Ultra NVL576。互联GPU数将从72颗进一步向576颗发展。届时，英伟达将在新 一代Kyber机架架构中引入NVLink Switch Blade(NVLink交换机刀片)，通过PCB中板替代传统5000+根有 源铜缆。可以看到，Rubin Ultra NVL576仍具有较强的工程创新能力。 Hopper架构开启超节点Scale up初步探索。GH200通过NVLink和NVLink-C2C(Chip-to-Chip)技术，使得 每个GPU可以访问其他所有CPU和GPU芯片的内存，实现GPU与CPU内存统一编址。 Blackwell架构推动Scale up标准化。GB200 NVL72将Scale-up规模稳定在72 个 GPU/机柜，形成可复制 标准化方案。NVL72由 18 个 Compute Tray(计算 ...

超节点与 Scale up 网络专题之英伟达： 行业标杆，领先优势建立在 NVLink 和 NVLink Switch 2026 年 2 月 5 日 看好/维持 通信 行业报告 | | | 投资摘要： 大语言模型（LLM）参数规模从千亿级向万亿级乃至十万亿级演进，跨服务器张量并行（TP）成为必然选 择；此外混合专家（MoE）模型在 Transformer 架构 LLM 中的规模化应用，更使跨服务器专家并行（EP） 成为分布式训练和推理的关键技术需求。为应对 TP 和 EP 对网络带宽与延迟的极为严苛的要求，构建超高 带宽、超低延迟的 Scale up 网络（纵向扩张网络）成为业界主流技术路径。 目前英伟达超节点已经推出成熟方案。2024-2026 年，英伟达陆续推出 GH200 NVL72、GB200/ GB300 NVL72、VR200 NVL72 三代超节点。 在超节点方案上，英伟达处于领先优势。2024-2025 年，英伟达陆续推出 GH200 NVL72、GB200/ GB300 NVL72 等成熟超节点解决方案。根据大摩预测，2025 年英伟达 GB200/300 NVL72 出货量约 2800 台 ...

据讯石光通讯网1月30日报道，Light Counting 发布最新报告指出，AI发展正推动光收发器和共封装光学 （CPO）市场高速增长。预计2025年市场规模达165亿美元，2026年将激增至260亿美元，年增长率高达60%。 尽管供应链短缺正在缓解，但顶级云公司的资本开支狂热为市场提供了支撑。Meta和Oracle计划在2026年将其 资本开支翻倍，其他公司尚未更新其2026年的支出计划。 有券商研究机构表示，根据当前的假设，2026年面向前五大云公司的光学器件销售额将占其资本开支的3.1% （高于2025 年的2.7%），并将在2031 年增至4.1%。光学连接支出增加可以归因于新的应用，例如Scale-up网 络，以及用于Scale-out 和Scale-up连接的GPU带宽的快速增长。CPO在Scale-up连接中的采纳可能会超出预测， 并在2028 年至2031年间引领市场实现更强劲的增长。 另外，英伟达2月3日举办共封装硅光子交换机相关研讨会NVIDIA将于2026年2月3日（新加坡时间14:00-15:00/ 澳大利亚东部时间17:00-18:00）举办网络研讨会，主题为"面向十亿瓦级AI工厂 ...

Santa Clara-based Nvidia hardly needs an introduction. Once celebrated as the king of gaming graphics, it quietly reinvented itself as the backbone of modern computing. Its GPUs now power data centers, AI, robotics, and immersive digital worlds. The CUDA software platform locked developers into a powerful ecosystem, turning Nvidia into an industry standard rather than a supplier. With a market capitalization of nearly $4.7 trillion, Jensen Huang’s company has become the engine of the AI economy.With Nvidia ...

如果您希望可以时常见面，欢迎标星收藏哦~ 通信技术的发展始终朝着更高速度的方向迈进——从电话线到光纤，从3G到5G。然而，一项名为 SerDes （串行器/解串器）的、已有数十年历史的技术，如今却在半导体行业引起了特别的关注。 这项成熟的技术为何突然成为热门话题？ 答案很简单：人工智能。 训练像 ChatGPT 这样的大型 AI 模型需要数千个 GPU 协同工作。挑战在于这些 GPU 必须交换的 海量数据。无论单个 GPU 的性能多么强大，如果数据在它们之间传输的速度不够快，整个系统就 会遇到瓶颈。 SerDes 是 这 条 " 数 据 高 速 公 路 " 背 后 的 技 术 。 随 着 人 工 智 能 对 带 宽 的 需 求 不 断 突 破 现 有 限 制 ， SerDes 已迅速从"锦上添花"的组件跃升为行业不可或缺的"关键技术" 。 SerDes 是Serializer和Deserializer的合成词。它的功能出奇地简单。 在计算机内部，数据沿着多条并行线路传输——就像一条八车道高速公路。然而，当在芯片或设备 之间传输数据时，维持这种并行结构就变得非常棘手。"八车道"需要大量的物理线路，而且要同步 所 ...

公众号记得加星标⭐️，第一时间看推送不会错过。 通信技术的发展始终朝着更高速度的方向迈进——从电话线到光纤，从3G到5G。然而，一项名为 SerDes （串行器/解串器）的、已有数十年历史的技术，如今却在半导体行业引起了特别的关注。这 项成熟的技术为何突然成为热门话题？ 答案很简单：人工智能。 训练像 ChatGPT 这样的大型 AI 模型需要数千个 GPU 协同工作。挑战在于这些 GPU 必须交换的海 量数据。无论单个 GPU 的性能多么强大，如果数据在它们之间传输的速度不够快，整个系统就会遇 到瓶颈。 SerDes 是这条"数据高速公路"背后的技术。随着人工智能对带宽的需求不断突破现有限制，SerDes 已迅速从"锦上添花"的组件跃升为行业不可或缺的"关键技术" 。 SerDes 是Serializer和Deserializer的合成词。它的功能出奇地简单。 在计算机内部，数据沿着多条并行线路传输——就像一条八车道高速公路。然而，当在芯片或设备之 间传输数据时，维持这种并行结构就变得非常棘手。"八车道"需要大量的物理线路，而且要同步所有 八条线路的传输时间也极其困难。 SerDes方案非常巧妙：在出发点将 ...

Core Insights - Upscale AI, a chip startup, has raised $200 million in Series A funding to challenge Nvidia's dominance in rack-level AI systems and compete with companies like Cisco, Broadcom, and AMD [1][3] - The rapid influx of investors reflects a growing consensus that traditional network architectures are inadequate for the demands of AI, which require high scalability and synchronization [1][2] Funding and Market Position - The funding round was led by Tiger Global, Premji Invest, and Xora Innovation, with participation from several notable investors, bringing Upscale AI's total funding to over $300 million [1] - The AI interconnect market is projected to reach $100 billion by the end of the decade, prompting Upscale AI to focus on this growing sector [6] Technology and Product Development - Upscale AI is developing a chip named SkyHammer, optimized for vertical scaling networks, which aims to provide deterministic latency for data transmission within rack components [9][10] - The company emphasizes the importance of heterogeneous computing and networks, believing that no single company can provide all the necessary technologies for AI [10][12] Competitive Landscape - Nvidia's networking revenue has seen a significant increase, with a 162% year-over-year growth, highlighting the competitive pressure in the AI networking space [3] - Upscale AI aims to create a high radix switch and a dedicated ASIC to compete with Nvidia's NVSwitch and other existing solutions [14][16] Strategic Partnerships and Standards - Upscale AI is building its platform on open standards and actively participating in various alliances, including the Ultra Accelerator Link and SONiC Foundation [7][17] - The company plans to expand its product line to include more traditional horizontal scaling switches while maintaining partnerships with major data center operators and GPU suppliers [18]

Core Insights - Nvidia dominates the data center GPU market and is expanding into networking, particularly with its rack-scale AI solutions [1] - In Q3 fiscal 2026, Nvidia's networking revenue reached $8.2 billion, a 162% increase year over year, driven by demand from major companies building AI data centers [2] - The networking attach rate for Nvidia's AI systems is nearly 90%, indicating strong integration of networking products with AI solutions [4] Networking Growth - The networking requirements for AI data centers differ significantly from standard cloud data centers, necessitating high data throughput for GPU efficiency [5] - Nvidia has captured an 11.6% share of the data center Ethernet switch market, ranking third behind Arista Networks and Cisco Systems [6] - The introduction of the Rubin platform, which integrates GPUs, CPUs, and networking technologies, is expected to enhance networking revenue significantly in 2026 and beyond [7][8] Customer Demand - Nearly 90% of customers purchasing Nvidia AI systems also acquire networking products, highlighting the growing demand for ultra-fast networking gear in large-scale AI data centers [9]

Core Viewpoint - Bank of America identifies a market setup linked to Nvidia that presents investment opportunities in AI infrastructure stocks rather than just focusing on Nvidia itself [1][4]. Group 1: Nvidia's Market Impact - Nvidia has seen a stock increase of over 1,082% in the past three years and over 30% in the last year, indicating its significant role in the AI trend [2]. - The company is effectively setting the pace for the entire AI industry, influencing the strategies of hyperscalers, server manufacturers, and data-center builders [4][5]. Group 2: Investment Strategy - Investors are advised to focus on industrial stocks that support the AI infrastructure rather than chasing Nvidia's stock movements [3][8]. - The market is currently overreacting to superficial signals, while the build cycle for AI infrastructure is expected to remain stable for the next few years [9]. Group 3: Nvidia's Ecosystem - Nvidia's CUDA ecosystem includes over 6 million developers and approximately 6,000 applications, creating a strong lock-in effect for companies that adopt its technology [6][7]. - Analysts estimate Nvidia's market share in AI to be around 80% to 85%, highlighting its dominance in the sector [7]. Group 4: Industrial Partnerships - Major industrial players are forming partnerships with data centers to adapt to Nvidia's ongoing redesigns, which presents a significant investment opportunity [8].

Core Viewpoint - The article emphasizes the critical importance of reliability in data centers, automotive, and aerospace industries, highlighting that failures can lead to significant economic impacts and potential loss of life [1]. Data Center Reliability Standards and Strategies - Cloud service providers operate hundreds of large data centers interconnected by thousands of miles of fiber optics, designed for high reliability with uptime ranging from 99.9% (43 minutes downtime monthly) to 99.999% (26 seconds downtime monthly) [2]. - Redundancy is key in data center design, with systems in place for load transfer and backup components to ensure continuous operation even during failures [2]. - Data centers utilize redundant cooling and power distribution systems to maintain operations during outages, with automatic switches to backup power sources [2]. Semiconductor Reliability Strategies - Data center chips must be designed for high reliability, employing fault-tolerant architectures to mitigate failures [3]. - Error-Correcting Code (ECC) memory is used in CPUs to enhance reliability, with advanced memory types like HBM3 incorporating stronger error correction methods [3]. - NVLink technology allows for low-latency communication between GPUs, with redundancy built into the system to maintain performance during component failures [5]. Component Design for High Reliability - Components are designed to detect early signs of failure and prioritize repairs, with redundancy to quickly identify and address issues [4]. - Modular and hot-swappable designs are encouraged to minimize downtime during component replacements [8]. Mechanical Engineering and Reliability - Mechanical engineering plays a crucial role in data center reliability, with the integration of multiple chips on a substrate posing risks of physical connection failures due to thermal and material differences [9]. - The operational temperature limits for data center components are significantly lower than those for automotive applications, with GPUs and processors designed to operate efficiently within these constraints [10]. Lifespan and Reliability Data - Data centers typically have a shorter lifespan of 5 to 6 years compared to automotive components, necessitating rapid deployment of new technologies [11]. - Extensive reliability data and stress testing are required before deploying new semiconductor components in data centers to ensure low failure rates [11]. Conclusion - High reliability in semiconductor architecture, firmware, and design is essential for success in the data center market, which is currently the largest segment for semiconductors [12].