文 | 硅谷101 我们的三小时新年直播来啦，在这场硅谷一线的思辨中，我们邀请了六位跨越AI、自动驾驶与美股投 资的领军人物，进行了一场长达三小时的直播讨论，聚焦聊聊2026年大家最关心的几个科技赛道展望。 这不只是一次预测，更是一场对"非共识"拆解。 AI范式的"去中心化"：2025年的"DeepSeek Moment"标志着大模型不再被五大厂垄断。2026年， 算力不再是唯一门槛，数据策展与系统级Scaling成为胜负手。 "自研模型"vs"应用优先"的大厂博弈：Meta的路线之争反映了科技巨头的战略分歧。在我们的讨 论中，张璐就认为底层模型技术是AGI时代的"电力"，必须自主掌控以确保独立性；Howie则主 张模型保鲜期短，大厂应发挥应用层优势，而非深陷基建泥潭。 端到端（E2E）的降维打击：旧金山停电事件中，特斯拉能够正常运营，而Waymo不知所措引发 全城大拥堵，特斯拉的端到端方案优势显现：在复杂的物理世界，基于规则的"有轨电车"方案正 在面临挑战。 算力版图的因子重构：美股投资正从"纯GPU信仰"转向"ASIC效率"。谷歌TPU阵营与英伟达的对 抗，本质上是推理成本与通用性之间的博弈。 信仰高于估 ...

智通财经APP获悉，有媒体援引知情人士透露的消息报道称，"AI芯片超级霸主"英伟达(NVDA.US)的最强劲竞争对手之一 ——人工智能芯片供应商Cerebras Systems Inc.正在商讨进行一轮大约10亿美元的新融资，以支持这家AI芯片初创公司与英 伟达之间的长期竞争态势，并且力争大幅强化该公司AI算力集群相比于英伟达AI GPU集群的性价比与能效比。 这位知情人士表示，这一轮融资将在投资前将这家初创公司的估值定在220亿美元，意味着较去年9月进行融资时的估值大 幅扩张170%。因为讨论尚属私下，该人士要求不具名。该知情人士还表示，这家AI芯片初创公司仍计划积极推进在美股的 首次公开募股(IPO)。 在首席执行官安德鲁·费尔德曼(Andrew Feldman)的带领之下，Cerebras Systems 正积极寻求挑战英伟达在人工智能芯片领域 的高达90%市场份额的绝对市场主导地位，因此这轮融资也势必将为寻求挑战全球最顶级人工智能算力基础设施供应商英 伟达主导地位的这家AI芯片公司带来新的一轮资金支持。 向英伟达下战书的Cerebras Systems是何方神圣? 英伟达前不久与AI芯片初创公司Gro ...

李恒春/文 来源：@证券市场周刊微博 读书介绍： 耐心是数字经济时代成长投资的必备品质，数字经济的产业变革需要时间，优质企业的价值释放也需要时间，那 些急于求成、追求短期收益的投资者，很难在成长投资中获得成功。 优质企业的成长需要时间，不可能一蹴而就。谷歌TPU的研发历经十年才实现突破，"易中天"组合的股价翻倍也 用了两年时间，美光的股价三倍上涨更是建立在多年的技术积累和产能布局之上。 亚当·西塞尔在《价值投资3.0：数字经济时代如何寻找百倍成长股》（以下简称《价值投资3.0》）中反复强调， 数字经济时代的成长投资，更需要坚守长期主义——"数字经济的产业变革需要时间，优质成长股的价值释放也需 要时间，短期的市场波动无法改变企业的长期成长趋势，反而会为长期投资者提供买入机会。" 2025年中美科技股的走势充满了波动与分化，谷歌、英伟达、美光以及A股"易中天"（新易盛、中际旭创、天孚 通信）组合的股价并非一路上涨，而是在市场的质疑与恐慌中不断前行。能够在这场波动中持续获利的投资者， 往往具备坚定的长期主义信念，能够穿越短期市场波动，坚守优质企业，这与西塞尔倡导的长期投资理念高度契 合。 拒绝短期噪音 坚守核心逻辑 ...

2026年1月6日开幕的CES 2026，老黄又宣称90%的ASIC项目会失败，这实际上是对此前谷歌TPU为代表的ASIC芯片（专用集成电路）的口 头"讨伐"，一场针对ASIC的全面围猎已经悄悄开始。 很多人会关心，GPU、ASIC竞争的终局如何？答案是取决于半导体战争的终极弹药库——台积电CoWoS先进封装产能。 这意味着，只要对台积电CoWoS产能预订、分配情况，进行颗粒度拆解，就能精确测算出2026年AI算力芯片的出货格局。 "未来六个季度数据中心收入5000亿美元。"黄仁勋在GTC25上说。 可以说，2026年"芯片战"，系于台积电115万片CoWoS晶圆产能。 GPGPU与ASIC阵营对垒，图片由AI生成 01 战争的起源 我们先对GPU和ASIC的战争背景做一些铺垫（有行业基础可跳过本部分）。 人工智能对算力的需求扩张是共识，但必须明确：更先进的计算架构、工艺制程和先进封装，是三个关键路径。 关于架构，谈到最多的是GPGPU（通用图形处理器），英伟达在这条路上，借助CUDA生态的20年铺垫，成为通用并行计算的绝对王者。 硬件层面，英伟达的核心武器有两个：HBM内存极高的带宽、GPGPU大规模流处理 ...

Group 1 - The capital market continues to be influenced by AI bubble theories, but significant advancements in model capabilities have been observed, particularly with Gemini 3, which surpasses ChatGPT in various evaluations, especially in "multimodal interactive" capabilities [3][45] - The AI industry is experiencing a competitive landscape where companies like OpenAI, Meta, and XAI are racing to enhance their models, with OpenAI planning to release GPT 5.3 in early 2026 to regain its leading position [4][46] - The competition has led to a shift in the tech industry, where companies are increasingly undermining each other rather than collaborating, as seen with OpenAI's entry into advertising and e-commerce, and Google's integration of AI into its search engine [5][47] Group 2 - In 2025, AI capabilities have evolved significantly, with reasoning becoming standard across major language models, and the cost of processing tokens decreasing by 50% [9][50] - Long-term memory capabilities have emerged in AI models, allowing them to remember user interactions and improve task execution strategies, which is essential for developing personal assistant applications [10][50] - The concept of "craft intelligence" has developed, where AI is expected to deliver satisfactory results in various tasks, reflecting a shift from merely providing accurate answers to replicating human best practices [11][51] Group 3 - The economic value generated by AI is complex, with significant investments in AI data centers (AIDC) expected to reach nearly $500 billion in 2025, leading to substantial depreciation costs for companies [15][16] - The revenue generated from AI applications is difficult to quantify, as it is spread across cloud vendors and enterprises that utilize AI tokens for internal improvements [17][19] - Companies are increasingly purchasing AI applications rather than building them in-house, with 76% of enterprises opting for external solutions in 2025, indicating a rapid acceptance of AI applications in the market [19][21] Group 4 - The future of AI applications is expected to bring transformative changes, including significant improvements in model performance and the potential for traditional software paradigms to be disrupted [23][25] - The integration of multimodal capabilities in AI models is anticipated to redefine content creation, moving towards an "experience industry" where video and interactive content become prevalent [32][34] - The demand for computational power in AI is projected to grow exponentially, with GPU and TPU technologies competing for dominance in the market [36][38]

技术垄断与封闭性。美国领先AI模型（如OpenAI的GPT-4、Anthropic的Claude）均采用闭源策略，核心技术与训练数据不对外开放。这种模式虽能保持短 期技术领先，但限制了研究透明度与社区创新，形成"技术孤岛"。 算力与硬件依赖。美国通过英伟达GPU、谷歌TPU等硬件垄断，构建了"硬件-框架-模型"的技术壁垒。例如，英伟达CUDA生态成为AI研发的事实标准， 导致其他国家难以突破算力瓶颈。 应用场景的局限性。美国AI应用多集中于消费互联网（如ChatGPT、MidJourney），对复杂工业场景、社会治理场景的渗透深度不足，难以解决真实世界 的"非结构化问题"（如工业设备故障预测、城市交通拥堵治理）。 中国AI发展跳出了"技术崇拜"的形式化陷阱，以"应用反哺技术"为核心逻辑，通过"场景驱动+全栈协同+生态构建"，形成了人、机、环境相互协同的智能 生态体系。其核心突破包括： 1. 从"技术定义需求"到"需求定义技术" 中国拥有全球最复杂、最多元的实体经济与社会治理场景（如工业制造、智慧交通、乡村振兴、医疗健康），这些场景成为AI技术的"试验场"与"试金 石"。例如： 工业场景，首钢股份冷轧公司落地6 ...

随着2026年的开启，全球市值最高的公司在股市中的根基正显露出动摇迹象。 自10月29日创下历史新高以来，英伟达(NVDA.US)股价已累计下跌9.1%，表现远逊于标普500指数。这反映出投资者日益担忧人工智能(AI)支 出的可持续性，以及这家芯片巨头的市场主导地位。 近期的下跌尤为引人注目。回顾收盘价刷新历史纪录当日，英伟达股价较2022年底涨幅已超过1300%，公司市值突破5万亿美元，较不到三年 前的约4000亿美元大幅跃升。然而如今，短短数月内其市值已蒸发4600亿美元，致使三年累计涨幅收窄至近1200%。 与此同时，这家占据主导地位的AI芯片制造商正面临前所未有的激烈竞争，对手不仅包括AMD(AMD.US)等同行，甚至还有其最大的客户——如 谷歌(GOOGL.US)和亚马逊(AMZN.US)等。更令华尔街忧心的是，英伟达对多家客户的投资行为，可能被解读为通过人为手段提振需求。 "风险显然已经上升，"资产管理规模达1300亿美元的Advisors Capital Management合伙人兼投资组合经理JoAnne Feeney评论道。 如果英伟达继续延续这一颓势，将对大多数股票投资者产生影响。汇编数据 ...

Core Viewpoint - The article discusses the fluctuations in the artificial intelligence sector on the ChiNext board, highlighting the performance of various stocks and ETFs related to AI and optical modules, with a focus on investment opportunities in the context of increasing demand for computing power and AI applications [1][3]. Group 1: Market Performance - On January 6, the ChiNext artificial intelligence sector experienced a decline, with notable pullbacks in optical module CPO concept stocks such as New Yi Sheng and Zhong Ji Xu Chuang, both dropping over 3% [1]. - In contrast, storage chips and AI application stocks showed strong performance, with Beijing Junzheng leading gains at over 8%, followed by Tonghuashun with over 7%, and several other stocks like Zhongke Chuangda and Tongniu Information rising over 3% [1]. - The ChiNext artificial intelligence ETF (159363) saw early morning fluctuations, initially dropping over 2% but later narrowing the decline to 0.49%, with a trading volume exceeding 700 million yuan and net subscriptions surpassing 10 million shares [1][2]. Group 2: Investment Opportunities - A report from Dongwu Securities highlights that by 2026, the overseas computing power chain is expected to maintain an upward trend, benefiting from the demand for optical interconnects driven by both overseas and domestic computing power needs [3]. - The high-end optical module market is projected to grow annually, with increasing visibility of demand extending further into the future, which will support rapid growth in the performance of supply chain companies [3]. - The report emphasizes three key investment directions: 1. High-certainty leading stocks that are integrated into the domestic and overseas computing power supply chain, which are expected to benefit significantly from the rapid growth of the optical interconnect industry [3]. 2. Critical material positioning in the supply chain, as the demand for high-speed optical modules will lead to significant growth in upstream optical chips and components [3]. 3. Potential new players in the supply chain, as increased demand for various products and customer needs may allow more suppliers to enter the core supply chain of major North American CSPs [4]. Group 3: ETF and Index Performance - The ChiNext artificial intelligence index achieved a remarkable annual growth of 106.35% in 2025, significantly outperforming other AI-themed indices [4]. - The first ChiNext artificial intelligence ETF (159363) is recommended for investment, focusing on leading optical module companies and capturing AI thematic market trends, with over 70% of its portfolio allocated to computing power and over 20% to AI applications [5].

Core Viewpoint - The AI sector in the ChiNext market is experiencing volatility, with significant fluctuations in stock prices, particularly in the optical module CPO concept stocks, while storage chips and AI applications show active performance [1][8]. Group 1: Market Performance - On January 6, the ChiNext AI index saw a decline, with notable drops in optical module CPO stocks like New Yi Sheng and Zhong Ji Xu Chuang, both falling over 3% [1][8]. - Conversely, stocks in the storage chip and AI application sectors performed well, with Beijing Junzheng leading with an increase of over 8%, followed by Tonghuashun with over 7%, and several others rising more than 3% [1][8]. - The ChiNext AI ETF (159363) experienced a morning drop of over 2%, which later narrowed to a decline of 0.49%, with a trading volume exceeding 700 million CNY and net subscriptions surpassing 10 million units [1][8]. Group 2: Investment Opportunities - A report from Dongwu Securities highlights that by 2026, the overseas computing power chain is expected to maintain upward momentum, benefiting from both overseas and domestic demand for optical interconnects [3][10]. - The high-end optical module market is projected to grow annually, with increasing visibility of demand extending further into the future, which will drive rapid growth in the performance of supply chain companies [3][10]. - The report emphasizes three key investment directions: 1. High-certainty leading stocks that are integrated into the domestic and overseas computing power supply chain, expected to benefit from the rapid growth of the optical interconnect industry [3][10]. 2. Critical material positioning in the supply chain, where the demand for high-speed optical modules will lead to significant growth in upstream optical chips and components [3][10]. 3. Potential new players in the supply chain, as increased demand for speed and products will allow more suppliers to enter the core supply chains of major North American CSPs [3][10]. Group 3: Index Performance - The ChiNext AI index achieved a remarkable annual growth rate of 106.35% in 2025, significantly outperforming other AI-related indices [4][11]. - The index's performance over the years from 2021 to 2025 shows fluctuations, with annual returns of 17.57%, -34.52%, 47.83%, 38.44%, and 106.35% respectively [12][11]. - The ChiNext AI ETF is designed to track this index, focusing over 70% of its portfolio on computing power and over 20% on AI applications, effectively capturing the AI theme market [12][6].

英伟达以200亿美元现金，收购初创公司Groq的核心资产，这笔交易不仅刷新了英伟达自身的历史收购纪录，远超其2019年以70亿美元收购Mellanox的 规模，更以雷霆之势，将一颗冉冉升起的英伟达挑战者直接纳入麾下。 此次远非一次简单的收购，而是一场精心策划的战略围剿。毕竟已经冒出来一家谷歌TPU了，英伟达不允许再有任何真正的挑战者存在，要么被击败， 要么被收购。 目前，谷歌正积极接触那些依赖租赁英伟达芯片的小型云服务提供商，劝说它们在数据中心托管谷歌的TPU处理器。首战告捷，谷歌已与伦敦的 Fluidstack达成协议，将在纽约的一个数据中心部署TPU。 同时，谷歌的触角伸向了为OpenAI建造数据中心的Crusoe，以及向微软租赁芯片并与OpenAI签有供应合同的CoreWeave——后者被业内视为英伟达的 亲儿子。这一系列动作，意图再明显不过：谷歌不仅要抢夺英伟达的客户，更要直接切入其最核心、利润最丰厚的AI算力供应链。 0 1 谷歌的决心 谷歌为这一战所做的铺垫，远比我们想象的更为漫长，也更为深远。 早在2015年，第一代TPU v1的性能优势，就已经在AlphaGo的演进中得到了最戏剧化的体现： 1 ...