"龙虾时刻"引爆AI Agent！ 今日早盘，锦鸡股份突然20%涨停，网宿科技一度也是20%涨停，美利云开盘一字板。与此同时，AI Agent项目Clawdbot刷屏，上线不足一个月，GitHub星标数狂飙至5.8w+，单日环比暴涨62%。Clawdbot 因此被称为人工智能的"龙虾时刻"。 值得注意的是，谷歌云正式官宣涨价的消息也在此时传出。有消息指，谷歌将于2026年5月1日对 GoogleCloud、CDN Interconnect 、Peering 与AI与计算基础设施服务涨价，其中北美原地翻倍，全球云 计算涨价趋势迎来重大验证。 两大利好 数据中心板块迎来两大利好。 一是谷歌云于当地时间1月27日宣布，自2026年5月起将正式上调全球数据传输服务价格，其中北美地区 费率较当前水平提高约一倍。在人工智能算力需求持续激增、硬件及能源等上游成本结构性上升的背景 下，主流云服务商正逐步将系统性成本压力向终端传导，行业长期以来的降价趋势出现关键转折。 本次调价覆盖互联网络传输（CDN Interconnect）、直连对等（Direct Peering）及运营商对等（Carrier Peering）三类服务。具 ...

"龙虾时刻"引爆AI Agent！ 今日早盘，锦鸡股份突然20%涨停，网宿科技一度也是20%涨停，美利云开盘一字板。与此同时，AI Agent项 目Clawdbot刷屏，上线不足一个月，GitHub星标数狂飙至5.8w+，单日环比暴涨62%。Clawdbot因此被称为人 工智能的"龙虾时刻"。 值得注意的是，谷歌云正式官宣涨价的消息也在此时传出。有消息指，谷歌将于2026年5月1日对 GoogleCloud、CDN Interconnect 、Peering 与AI与计算基础设施服务涨价，其中北美原地翻倍，全球云计算涨 价趋势迎来重大验证。 两大利好 数据中心板块迎来两大利好。 一是谷歌云于当地时间1月27日宣布，自2026年5月起将正式上调全球数据传输服务价格，其中北美地区费率较 当前水平提高约一倍。在人工智能算力需求持续激增、硬件及能源等上游成本结构性上升的背景下，主流云服 务商正逐步将系统性成本压力向终端传导，行业长期以来的降价趋势出现关键转折。 本次调价覆盖互联网络传输（CDN Interconnect）、直连对等（Direct Peering）及运营商对等（Carrier Peering）三类服务。 ...

Core Viewpoint - China has approved the import of its first batch of Nvidia's H200 artificial intelligence chips, indicating a shift in its stance as it aims to balance AI needs with domestic development [1] Group 1 - The approval of Nvidia's H200 chips marks a significant change in China's policy regarding AI technology imports [1] - This decision reflects China's strategy to enhance its AI capabilities while also promoting local industry growth [1]

目前主要仍由英伟达垄断的人形机器人"大脑"芯片模组，迎来了国产替代的新选择。 1月26日，国产GPU企业天数智芯发布了芯片四代架构路线图，并推出"彤央"品牌系列边端侧AI算力 产，可用于具身智能，以及交通智能、工业智能、车路云一体化等典型边端场景。 彤央系列产品的突出特点在于兼容CUDA生态，机器人操作系统、具身模型及算法可在彤央产品和英伟 达Orin系列产品上丝滑切换，给具身智能领域的产业应用提供更多选择。 具体来看，此次发布的"彤央"系列包括四款产品，各有侧重：彤央TY1000算力模组采用 699pin 接口， 以口袋大小集成行业级算力与开放生态，实现便携化部署；彤央TY1100算力模组集成ARM v9 12核 CPU与自研GPU模组，以充沛算力提供多元选择；彤央TY1100_NX 算力终端凭借更大显存成为高性价 比之选，堪称边端算力"小钢炮"；彤央TY1200 算力终端则以 300 TOPS的极致性能与小巧身材，为 AIPC、具身智能等前沿场景提供核心支撑。 | 参数 | TY1200 | | --- | --- | | CPU | 16-Core X86 | | GPU | 天数智芯GPGPU MXM ...

硅谷与东京之间，一场关于人类文明算力上限的豪赌正进入博弈阶段。 1月28日，据知情人士披露，软银集团正处于洽谈向OpenAI追加高达300亿美元投资的后期阶段。这笔 资金不仅将使OpenAI的总融资规模向1000亿美元靠拢，更可能将这家人工智能领军企业的估值推向令 人咋舌的8300亿美元。 然而，这不仅仅是一次简单的股权更迭。在这笔巨额资本流动的背后，隐藏着一个由软银董事长孙正 义、OpenAI首席执行官山姆·奥特曼以及代工巨头鸿海共同编织的宏大蓝图：通过"星门计划"（Project Stargate）的基础设施工程，将AI从单纯的算法竞争转变为一场关于能源、土地与钢铁的重工业竞赛。 01 软银将与微软分庭抗礼 软银在2025年11月披露，该公司在2025年10月期间已清仓其持有的58亿美元英伟达股份。在外界看来， 这无异于在金矿繁荣期卖掉了最大的"锄头"供应商。 但对孙正义而言，逻辑清晰而决绝：他正将筹码从"通用硬件"转移到"超级智能"及其配套的垂直整合体 系中。 孙正义的这一转变反映了他对AI价值链重构的判断。他认为，尽管英伟达目前统治着芯片市场，但未 来的核心利润将向拥有"模型主权"和"算力主权"的公司 ...

来源：半导体产业纵横 这能否让我们摆脱当前硅技术的局限性？ 总部位于美国德克萨斯州奥斯汀的人工智能芯片企业Neurophos，背靠比尔·盖茨旗下盖茨前沿基金的资 本支持，近日公布了其光学处理单元（OPU）的核心测试成果。该公司声称，这款自研OPU在 FP4/INT4低精度计算工作负载下，算力性能达到英伟达最新款Vera Rubin NVL72人工智能超级计算机的 十倍，且二者功耗水平基本持平。据科技媒体The Register报道，Neurophos实现这一技术突破的核心路 径，在于采用了更大规模的计算矩阵与更高的运行时钟频率，这两大设计革新成为其光学计算芯片对标 传统GPU的关键优势。 Neurophos首席执行官Patrick Bowen在接受媒体采访时，详解了这款OPU的核心硬件设计亮点。他表 示，该芯片仅集成了一枚1000 x 1000像素规格的光子传感器，单从尺寸来看，这一矩阵规模约为当前多 数AI GPU所采用的256 x 256像素矩阵的15倍，更大的矩阵直接提升了芯片并行计算的基础能力。而更 具突破性的是，Neurophos成功将光晶体管的物理尺寸缩小了约10000倍，这一微型化成果解决了传统 ...

南方财经 21世纪经济报道记者吴斌 上海报道 在科技股和芯片股的提振下，标普500指数创下历史新高。 美国东部时间1月27日，标普500指数收涨0.41%，报6978.60点，时隔半个月再创历史新高；纳斯达克综 合指数涨0.91%，报23817.10点；道琼斯指数跌0.83%，报49003.41点。 财报方面，约五分之一的标普500成分股本周公布季度业绩，"七巨头"中有苹果、微软、Meta和特斯 拉。 本轮财报季的一个关键主题在于：企业是否开始真正从AI相关投资中获益？在2025年末，市场一度担 心大规模投向数据中心及其他基础设施的资本支出能否带来回报，这种疑虑一度打压科技股及其他AI 概念股，本周市场将对此高度关注。 科技巨头能否继续走高？ 1月27日，"七巨头"中六家上涨，英伟达、苹果涨逾1%，微软、亚马逊涨逾2%，仅特斯拉一家下跌。 Globalt Investments高级投资组合经理Thomas Martin分析称，大家都在关注任何能让外界深入了解人工 智能技术发展趋势的信息，投资者将重点关注各公司的资本支出水平以及与AI商业化相关的信息，除 了资本支出和运营支出总额之外，相关评论也将是关注的焦点。 ...

January 27, 2026 10:15 PM GMT Asia-Pacific Technology | Asia Pacific AI Supply Chain: TPU/ASIC updates; ICMS NAND demand calculation We sense that AI semi vendors have started securing critical 2027 components – T-Glass/ABF, HBM, and TSMC 3nm – (e.g., MediaTek's 3nm TPU). Downgrade Egis to EW, as its 2026 appears to be shaping up as a year of transition. ASIC – Volume upside for MediaTek's 3nm TPU project in 2027: In our Target Price Up report, we highlighted the supply chain's bull case of 6-7mn TPU units ...

Summary of Key Points from the Conference Call Industry Overview - The focus is on the U.S. Semiconductors industry, particularly the AI value chain and its implications for various companies involved in hardware and semiconductor verticals [1][2] Core Insights and Arguments - A top-down framework is constructed to dimension the AI upside across sectors from 2025 to 2027, aimed at simplifying comparisons across different companies [2][19] - The analysis updates the breakdown of data center capital expenditures (capex) to reflect rising prices for DRAM and NAND, estimating an increase of approximately $70,000 per rack for server DRAM and $35,000 per rack for storage costs, raising all-in capex from $5.9 million to $6.0 million per rack [3][21] - Incremental revenue estimates per gigawatt (GW) of capacity are derived from market share estimates across nine key hardware/semiconductor verticals, with a regression analysis used to estimate margins on incremental AI revenue [4][23] Company-Specific Insights - **Unimicron and Ibiden** are highlighted as having significant upside potential, particularly in ABF substrate and HDI, with Unimicron expected to benefit from large opportunities [5][27] - **Nvidia (NVDA)** and **Broadcom (AVGO)** are identified as industry favorites, with Nvidia's AI substrate upgrade expected to double content generation [5][27] - **Intel (INTC)** and **Cisco** are noted to have lower exposure to AI opportunities compared to their market prominence, with Intel facing challenges in capturing market share [6][28][31] - **Delta Electronics** is rated as outperforming, with a price target of NT$1,300, benefiting from increased electrical content in AI data centers [13][30] - **MediaTek** is expected to see growth from the TPU ramp, while memory/storage players like **SanDisk**, **Samsung**, **Micron**, **SK Hynix**, and **KIOXIA** are projected to benefit from rapid memory price surges [6][30] Financial Metrics and Valuations - The report includes a detailed ticker table with performance metrics for various companies, including EPS and adjusted P/E ratios, indicating that Unimicron has room for growth while Intel appears expensive relative to its AI opportunities [9][29] - **Nvidia** is rated outperform with a target price of $275, while **AMD** is rated market perform with a target of $225, reflecting high expectations for AI growth [11][12] Additional Considerations - The analysis acknowledges that estimates of AI upside are imprecise and that valuations are influenced by various non-AI factors, suggesting that investors should consider their own assumptions for more accurate estimates [20][26] - The report emphasizes the importance of understanding the nuances of each company's position within the AI landscape, including their ability to adapt to evolving data center requirements [20][30] Conclusion - The semiconductor industry is poised for significant growth driven by AI, with specific companies like Unimicron and Ibiden positioned to capture substantial market share. However, challenges remain for established players like Intel, highlighting the dynamic nature of the sector and the need for careful analysis of individual company prospects [5][6][27][30]

CoreWeave | North America Deepening the Alliance - CoreWeave & NVIDIA Expand Partnership Key Takeaways What Was Announced? CoreWeave announced a multi-pronged expansion of its partnership with NVIDIA (see here) focused on 1) accelerating the build out of an incremental >5 GW of AI infrastructure by 2030 (CoreWeave-specific forecast, with NVIDIA likely a very small portion), 2) accelerating CoreWeave's ability to obtain land, power and shells to build these AI factories, 3) adding a new dimension of software ...