Key Takeaways NVIDIA targets $65B in Q4 revenues after posting a record $57.01B in Q3 fiscal 2026.The data center unit generated $51.22B in Q3, up 66% year over year.Analysts see NVIDIA reaching $65.56B in Q4, driven by AI chip and Blackwell demand.NVIDIA Corporation (NVDA) is set to report its fourth-quarter fiscal 2026 results on Feb. 25, and it would be no surprise if the company reaches another sales milestone and surpasses the $65 billion target. NVDA has been setting new sales records for the past few ...

中金研究 大模型的演进史，本质上是一部与"遗忘"抗争的历史。 当我们惊叹于模型的推理能力时，往往忽视了一个重要短板： 在缺乏记忆留存的架构下，模型 每一次对历史信息的处理，本质上都是一次昂贵的"重复计算"。 这种以高昂算力对抗遗忘的粗放模式，正面临着显存墙与上下文窗口的物理极限。我 们认为，2026年及之后的AI Infra主战场将增加"模型记忆"这一极。 何为模型记忆？如何理解短期、中期、长期记忆三层记忆系统对应的软硬件需求？ 如何对应模型训练、推理、Agent场景理解记忆分层系统？我们将在本报告中予以解答。 点击小程序查看报告原文 Abstract 摘要 短期记忆构成大模 型单 次推理的"当前视野"。 作为高频读写、对延迟极度敏感的"热数据"，其核心矛盾在于KV Cache对显存容量与带宽的双重挤占。软 件端通过PagedAttention显存虚拟化与PD分离调度进行优化，并探索出无限注意力（Infini-attention）等前沿架构以支撑百万Tokens上下文窗口。这一逻辑 直接锚定了HBM与片上SRAM作为突破"显存墙"与"延迟墙"的重要硬件要素。 中 期记忆保障跨会话的情景连续性，是Agent的基 ...

当谷歌的大模型 Gemini 3 在2025年末以惊人的多模态处理速度和极低的延迟震撼业界 时，外界往往将目光聚焦于算法的精进。然而，真正的功臣正沉默地跳动在谷歌数据中心 的机架上——那就是他们潜研10年的 TPU (Tensor Processing Unit)。 长期以来，英伟达凭借其"通用而强大"的 GPU 统治了模型训练的黄金时代。但随着大模 型 走 进 规 模 化 应 用 爆 发 期 ， 算 力 逻 辑 正 发 生 本 质 改 变 ： " 训 练 为 王 " 的 旧 秩 序 正 在 瓦 解，"推理为王"的新时代已经降临。 当专用架构的极致效率突破了通用架构的冗余局限，以 TPU 为代表的 ASIC 芯片正以不 可阻挡之势，从英伟达手中接过主角的剧本，重塑全球AI算力的权力版图。 成本为王，芯片变了 这些年，在海内外厂商的共同推动下，大模型和人工智能成为了几乎人尽皆知的热词。所谓大模 型，其诞生有点像一个人的成长：先通过预训练"博览群书"，在海量文本中学习语言结构和世界知 识；再通过指令微调，学会如何按人类要求组织和表达回答；接着借助基于人类反馈的强化学习， 对齐输出风格与边界，使回答更符合人类偏好； ...

Core Viewpoint - Starting from 2025, supernodes will become a significant technological innovation direction in the AI computing network, with increasing competition among AI chip manufacturers in both chip performance and Scale up network [1][5]. Group 1: Supernode Development - Nvidia has launched mature supernode solutions, with plans to release GH200 NVL72, GB200/GB300 NVL72, and VR200 NVL72 from 2024 to 2026 [1][3]. - The Blackwell architecture standardizes Scale up with GB200 NVL72 stabilizing the scale at 72 GPUs per cabinet, consisting of 18 Compute Trays and 9 Switch Trays [2]. - The Rubin architecture will enhance bandwidth, with the NVLink 6 Switch increasing single GPU interconnect bandwidth to 3.6 TB/s, up from 1.8 TB/s [2]. Group 2: Nvidia's Competitive Advantage - Nvidia maintains a leading position in the supernode market, with a projected shipment of approximately 2,800 units of GB200/300 NVL72 by 2025 [3]. - Future plans include the introduction of Vera Rubin NVL144 and Rubin Ultra NVL576, expanding interconnected GPUs from 72 to 576 [3]. - Innovations such as NVLink and NVLink Switch are crucial for achieving high bandwidth and low latency in AI training clusters, with NVLink 5 Switch supporting a total bandwidth of 130 TB/s for 72 GPUs [4]. Group 3: Industry Landscape and Investment Strategy - The global supernode competition landscape is still being established, with Nvidia currently in a leading position [6]. - The report suggests monitoring Nvidia's supernode supply chain, including components like PCB backplanes, high-speed copper cables, optical modules, and cooling systems [6]. - Chinese manufacturers are actively participating in the supernode and Scale up network sectors, with potential for domestic firms to gain a competitive edge [6].

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 ...

Core Insights - Nvidia has transformed from a gaming graphics company to a key player in modern computing, with a market capitalization of nearly $4.7 trillion, becoming essential to the AI economy [1]. Company Overview - Nvidia's GPUs are now integral to data centers, AI, robotics, and immersive digital environments, supported by the CUDA software platform that has established it as an industry standard [1]. - The company has a long-standing relationship with OpenAI, participating in a $6.6 billion funding round and committing up to $100 billion for data center expansion, with OpenAI leasing millions of Nvidia chips in a deal valued at hundreds of billions [2]. Financial Performance - Nvidia's revenue for fiscal Q3 reached $57.1 billion, a 62% year-over-year increase, with adjusted EPS climbing 60% to $1.30 per share [11]. - Data center revenue grew 66% year-over-year to $51.2 billion, while networking revenue surged 162% to $8.2 billion, and gaming revenue increased by 30% [12]. - Operating cash flow rose to $23.8 billion, and free cash flow increased by 65% to $22.1 billion, with $37 billion returned to shareholders in the first nine months of fiscal 2026 [13]. Market Position and Valuation - Nvidia's stock is trading at approximately 42.56 times forward adjusted earnings, which is above most peers but below its long-term average, indicating that the premium may be justified given strong earnings growth [9]. - Analysts maintain a consensus rating of "Strong Buy" for Nvidia, with an average price target of $254.81, suggesting a potential upside of 34% [16]. Future Outlook - Nvidia is expected to report revenue of around $65 billion for Q4, with analysts projecting a 70.6% year-over-year EPS growth to $1.45 [14]. - For the full fiscal 2026, EPS is projected to increase by 51.2% to $4.43, followed by a further rise of 58.9% to $7.04 in fiscal 2027 [15].

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]