Core Viewpoint - Nvidia's CEO Jensen Huang predicts that the company's AI processors will generate $1 trillion in sales by 2027, highlighting the increasing demand for computing power and Nvidia's capability to meet this demand [1][3]. Product and Technology Developments - Nvidia plans to expand into central processing units (CPUs), traditionally dominated by Intel, and introduced new semiconductors developed from technology acquired from startup Groq [2][16]. - The company is developing chips for data centers in outer space and has introduced the Groq 3 LPU, a specialized chip for accelerating large language model inference [2][12]. - Nvidia's upcoming CPU, branded Vera, is designed to handle complex tasks more efficiently and will require less electricity [17][16]. - The company is also launching computers made entirely of CPUs, marking a new approach for Nvidia [18]. Market Position and Financial Outlook - Nvidia's sales forecast for data center gear has been extended to $500 billion through 2027, doubling the previous estimate [4]. - Despite a strong sales growth, Nvidia's stock has faced volatility, with shares down 3.4% this year prior to the GTC presentation [11]. - The company remains the world's most valuable company, with a market value of $4.4 trillion, driven by a surge in AI chip spending [9][11]. Partnerships and Industry Impact - Nvidia announced new or expanded partnerships with major companies like IBM, Hewlett Packard Enterprise, and Adobe, aiming to showcase the benefits of AI across various industries [8][7]. - The company's entry into the CPU market may pose a challenge to Intel and increase competition with in-house chip efforts from companies like Amazon [23][21]. Future Innovations - Nvidia is set to release its next-generation AI processors, Vera Rubin, in the second half of 2026, followed by a design named after physicist Richard Feynman [10]. - The company is also enhancing its technology development cycle, aiming to replace its entire product lineup annually [10].

文 | 《硅谷观察》栏目 郑骏 硅谷圣何塞SAP中心，几乎完全坐满。 这座平时属于NHL圣何塞鲨鱼冰球队的主场，在今天变成了一年一度的"AI麦加"。来自190个国家的开 发者、工程师、企业买家和投资人挤满了每一个座位，所有人的目光都对准一个熟悉的身影：那个爱穿 皮夹克的中年男人。 英伟达CEO黄仁勋走上台的第一句话是："It all starts here．"——一切从这里开始。接下来的两个多小 时，他兑现了这句话的分量。他笑着说，今天这里就像是超级碗一样。 他预计英伟达新一代AI加速芯片架构Blackwell与下一代Rubin产品，到2027年底将创造至少1万亿美元 收入。这一数字远超黄仁勋2025年10月给出的5000亿美元销售预测，再次凸显AI基础设施投资浪潮仍 在快速扩张。 核心产品：Vera Rubin全栈亮相 Vera Rubin是本场演讲的绝对主角，只不过黄仁勋一直等到一个半小时之后才正式发布。这套系统在去 年年底的华盛顿特区活动就已经披露，今年年初的CES 2026进一步展示细节，而今天则是完整正式发 布。核心亮点如下： Vera Rubin NVL72是当前旗舰规格，由72颗GPU通过NVL ...

英伟达表示，Groq 3 LPX 是Vera Rubin的推理加速器，旨在满足智能体系统对低延迟和大上下文的需求。Vera Rubin和LPX通过协同设计 的架构，将Rubin GPU和LPU的卓越性能完美结合，使其呈现极低延迟与超大吞吐量的特点。 黄仁勋介绍称，LPX与Vera Rubin平台结合后，推理吞吐量/功耗比将能提升35倍。LPU芯片将由三星代工，预计机架将于今年下半年开始 出货。 在GTC 2026主题演讲上，一款名为英伟达Groq 3 LPU的全新芯片正式亮相。 就在昨日，分析师郭明錤发文称，在英伟达投资Groq之后， LPU的出货量预测已大幅上调 。预计2026至2027年的总出货量将达到400万 至500万颗。新架构机柜预计于今年第四季度开始量产，2026年与2027年机柜出货量分别约为300至500个，以及15000至20000个。 北京时间周二凌晨，英伟达正式推出Vera Rubin平台，包括Groq 3 LPU（简称LPU）在内，共计搭载了七款芯片，分别为Vera CPU、 Rubin GPU、NVLink 6交换机、ConnectX-9超级网卡、BlueField-4 DPU和Sp ...