【文/观察者网 阮佳琪】 他坦言，"我们必须承认，华为是全球范围内极具实力的科技企业之一。我们与该公司存在竞争关系， 他们实力强劲、反应敏捷且行动迅速。" 在周三的活动中，黄仁勋还提到，中国在AI竞赛中相对美国的另一大优势是能源供应充足。他说，"中 国的能源总量是美国的两倍，而美国经济规模却大于中国，这在我看来完全不合常理。" 基于此，他呼吁美国"必须利用一切可用的能源形式"，并强调美国"应该努力鼓励和加速"核能发展。 急，很急，特别急——这就是英伟达首席执行官黄仁勋近来的真实写照。为了搞定对华芯片出口，最近 一段时间他都在忙活着向各方强调事态"十万火急"。 据《日经亚洲》4日报道，当地时间周三，在华盛顿智库美国战略与国际问题研究中心（CSIS）举办的 一场活动上，黄仁勋第N+1次发出警告称，如果美国企业放任华为等中国竞争对手抢占市场，中国将很 快寻求向全球输出其人工智能（AI）技术。 活动上，黄仁勋表示，美国限制英伟达芯片对华出口的举措，"实质上已拱手让出全球第二大人工智能 市场"，这将为华为等中国本土企业的技术成熟提供发展空间，最终使其具备在全球层面与美国企业抗 衡的实力。 "中国市场的地位无可替代，我们 ...

FT：美国已停止了对中国国家安全部实施制裁的计划，此前该部门被指控发起了一场大规模的网络活动，此举旨在避免破坏特朗普和习近平在10月达成的贸易休战协议。根据几位美国官员和其他知情人士的说法，在釜山达成的协议之后，本届政府也不会对中国实施新的重大出口管制措施。几位知情人士表示，特朗普政府对华政策的目标已经转向确保“稳定”，直到减少中国在稀土领域的支配地位为止，稀土支配地位限制了美国采取激进行动的能力。特朗普也不想危及他定于4月对北京的访问。据知情人士透露，另一个引发对华鹰派担忧的举动是，本届政府正准备举行一次高级别会议，决定是否发放许可，允许英伟达向中国出口先进芯片H200。一位熟悉白宫想法的消息人士表示，特朗普“致力于与中国建立互利的贸易关系，同时不损害我们的国家和经济安全”。外汇交易员 (@myfxtrader):彭博：英伟达的强力游说取得关键成果，一项旨在限制其向部分海外市场销售先进AI芯片的《GAIN AI法案》，预计将不会被纳入美国年度国防法案。该法案原计划要求英伟达和AMD等芯片制造商，在向受限制国家出口其高性能AI芯片之前，必须优先满足美国客户的需求。 https://t.co/HMCGD6JQl ...

Nvidia um needing to sell chips to China. I guess the Jensen Hang is you really making all these trips to um to meet with lawmakers with the president. Do you think the idea is that yes, this is really the window of major need for China's selfsufficiency and the window if we don't do something about it, it will be opened even wider rather than closed.What do you think. >> Well, China has had a semiconductor self-sufficiency goal and plan as as uh as early as 2015. So this is 10 years in the making.Um I unde ...

Core Viewpoint - Nvidia has acquired a $2 billion stake in Synopsys, making it one of the largest shareholders in the chip design software company, which positively impacted the stock prices of both firms [1]. Company Summary - Nvidia's investment in Synopsys highlights its strategy to strengthen its position in the semiconductor industry by leveraging Synopsys' expertise in chip design software [1]. - The acquisition is expected to enhance Nvidia's capabilities in developing advanced technologies, particularly in AI and machine learning applications [1]. Market Reaction - Following the announcement, shares of both Nvidia and Synopsys experienced an increase, indicating positive market sentiment towards the investment [1].

另外，三星已完成HBM4的内部测试；如果包括英伟达在内的主要客户的质量认证结果最早在本月公布，三星可能开始量产。 三星之所以能在SOCAMM 2 的导入阶段抢下英伟达超过50%订单，可能在于其10nm级第6代（1c）DRAM在良率与性能表现较为出色，使得SOCAMM 供 应链格局出现明显调整，领先了本来占优势的美光。 资料显示，SOCAMM 是近年英伟达推动的高性能内存模块，采用压缩连接式CAMM（Compression Attached Memory Module）设计，能在更薄的基板上 整合多颗低功耗DRAM，较传统DDR5 RDIMM 在空间效率与能耗上更具优势。 ***************END*************** 国芯网[原:中国半导体论坛] 振兴国产半导体产业！ 不拘中国、 放眼世界 ！ 关注 世界半导体论坛 ↓ ↓ ↓ 12月4日消息，据韩媒报道，三星电子获得英伟达2026年SOCAMM 2芯片一半的订单！ 英伟达需要从全球DRAM行业采购约200亿GB的SOCAMM 2，三星获得约100亿GB的订单，SK海力士可能承担约60-70亿GB的订单，剩余部分由美光供 应。 第一步：扫描 ...

Group 1 - China's Aerospace Science and Technology Corporation has completed its first self-built commercial launch site in Jiuquan, marking a significant milestone in the country's commercial space endeavors [2] - Samsung Electronics has completed the development of its sixth-generation High Bandwidth Memory (HBM4) chips and is preparing for mass production, with samples sent to NVIDIA for quality testing [2] - The Zhuque-3 reusable rocket successfully launched but experienced an anomaly during the recovery phase, failing to achieve a soft landing; the team is conducting a thorough review to identify the cause and improve recovery strategies [2] Group 2 - The European Union is pushing for a "European manufacturing" requirement of 70% for key products, which could lead to an annual cost increase of over €10 billion for companies [2] - The Chengdu Woge project is set to enter the equipment debugging phase for the world's first 8.6-generation AMOLED glass processing line, expected to begin trial operations in January 2026, enhancing local supply chain efficiency [2]

RT Hans C Nelson 🗽 (@HansCNelson)Listen to Jensen Huang recount this story of a terrifying conversation he had to have with SEGA executive Shoichiro Irimajiri (aka "Iri") in the early 1990s after he realized that NVIDIA's failures were going to cost them everything.Jensen's clear thinking about the problems NVIDIA & Sega faced combined w/ his willingness to be humble, forthright, 100% honest, and vulnerable in that conversation w/ Iri, as well as Iri's good sense and generosity, saved NVIDIA from certain de ...

Core Insights - The recent advancements in AI models, particularly Google Gemini and Anthropic's Cloud, indicate an acceleration in AI progress, which is positive for the AI industry and companies like Nvidia [1] - OpenAI is expected to release a more efficient model in the coming months, leveraging the scaling techniques demonstrated by competitors [2] - The current AI market is not a bubble; instead, it is at the beginning of a multi-year demand acceleration phase [2][3] AI Model Advancements - Google Gemini and Anthropic's models have shown significant improvements in major benchmarks, suggesting a bullish outlook for AI adoption [1] - OpenAI is set to benefit from Nvidia's new NVL 72 AI server clusters, enhancing its model training capabilities [1] Market Valuation Comparison - AI stocks are currently trading at 25 to 30 times earnings, which are believed to be understated, contrasting sharply with the dot-com bubble where stocks traded at 100 times forward earnings [4] Data Center Buildout Cycle - The data center capacity leasing has seen an unprecedented increase, with more capacity leased in the September quarter than in the entire previous year [6] - Major hyperscalers, including Microsoft and Amazon, are planning to double their data center capacity over the next four quarters due to overwhelming demand [8] Investment Behavior - The investment behavior in AI is not reckless; companies are strategically building data center capacity in response to actual demand from startups and enterprises [8][9] - The Nvidia and OpenAI deal is still in the finalization stage, indicating that many anticipated investments have yet to materialize [9][10]

2025年12月4日,黄仁勋在华盛顿的一场访谈中,给AI产业竞赛重新标定了坐标系。 当主持人问及NVIDIA如何看待AI竞争格局时,黄仁勋的回答是: 如果你没有能源,你就没有AI。 不是大模型,不是AGI,不是算法创新。 他说,今天用于AI数据中心的GPU重两吨,消耗20万瓦,要价300万美元。运行一个真正的AI数据中心,你得 填满一个足球场。 AI 模型可以复制，AI 工厂不能。 这场产业竞赛的本质,不是谁的算法更强,而是谁建得更快、电力充足、工厂批量落地。 模型吞吐量、训练周期、推理规模,看上去是算法问题,实质是电力问题。 一台AI数据中心的GPU,是重达两吨、耗电20万瓦的工业设备。而要把这些GPU跑起来,不是买几台放机 房,而是要建出一整套工业级电力系统。 于是我们第一次清晰看见：AI竞争的真正战场不在硅谷，而在作为底座的基础设施。具体而言,是能 源、芯片、平台、建厂速度和资本调度能力。 谁能建出 AI 工厂,谁就是赢家。 第一节｜能源,才是AI工厂的底座 黄仁勋在访谈里,把能源放在了AI五层堆栈的最底端。 他不是在打比方,这是产业事实: 如果我们没有能源增长,这个产业就无法发展。 01｜为什么能源成 ...

21世纪经济报道新质生产力研究院综合报道 早上好，新的一天又开始了。在过去的24小时内，科技行业发生了哪些有意思的事情？来跟21tech一起 看看吧。 【巨头风向标】 豆包手机助手回应"侵犯隐私"：需用户主动授权，产品目前不面向普通消费者 12306回应"取消靠窗选座" 近日，有网友发帖称，在12306购票选座时发现，A、F座两侧的"窗"字没了，而是变为"两堵墙"。对 此，12月2日，记者致电12306客服电话咨询，工作人员回应称，最近对于购票选座界面进行了优化调 整，A、F座两侧并不是"墙"，而是将"窗"字改为一道竖线。A、F座依然是靠窗侧的座位，只是不一定 正对窗，可能在窗户前或窗户后，界面优化是为了避免因座位是否靠窗引起的误解。此前，有网友反映 其购买的高铁靠窗票，上车后却发现是没有窗户的"面壁座"，建议铁路部门细化座位标识，以便消费者 选择。 国产GPU独角兽摩尔线程将于12月5日在科创板上市 12月3日晚间，摩尔线程披露公告称，经上交所审核同意，公司发行的人民币普通股股票将于2025年12 月5日（本周五）在上交所科创板上市。公告显示，根据《上海证券交易所科创板上市公司自律监管指 引第5号——科创成 ...