此次上市对整个AI板块具有风向标意义。智谱不仅是中国独立通用大模型开发商中收入规模最大的企业，也是首个接受二级市场定价考验的基座 模型厂商。 市场反应表明，尽管面临短期盈利压力，但投资者更看重其在模型迭代速度、开发者生态构建以及政企落地能力上的稀缺性。 智谱成立于2019年，由清华大学计算机系知识工程实验室（KEG）转化而来，具备从底层算法到全链路自主研发能力。公司愿景是实现通用人工 智能（AGI），目前已构建了包括语言、视觉、代码及智能体在内的全栈模型矩阵。在资本热潮退去、行业回归理性的当下，智谱如何通过MaaS （模型即服务）模式平衡高强度的研发投入与商业回报，成为后续市场关注的核心。 作为自清华大学计算机系转化而来的中国人工智能独角兽，智谱AI正式登陆港交所，成为"国产大模型第一股"。 1月8日，智谱正式在港交所主板挂牌上市，发行价定为116.2港元。尽管开盘一度破发，但股价随后迅速回升并转涨，盘中一度涨超10％。截至发 稿时，智谱股价报130港元，涨幅12％，总市值达574亿港元。本次IPO香港公开发售部分获超1159倍认购，显示出零售投资者对这一稀缺AI标的 的高度热情，国际发售部分亦获逾15倍认购 ...

作为享誉全球的经济学家、国际发展领域权威，萨克斯以 "休克疗法" 经济理论闻名，曾助力多国化解债务危机与恶性通胀，兼具深厚学术功底与实战经 验。 萨克斯强调，自1947年以来，美国外交政策多次通过武力、秘密行动和政治操纵，在其他国家推动政权更迭。这些措施均违反《联合国宪章》，且通常导致 持续暴力、致命冲突、政治动荡，以及平民的深重苦难。 第一次世界大战后，国际联盟成立，旨在通过适用国际法终结这一悲剧，但世界主要国家未能在20世纪30年代捍卫国际法，最终引发了新的全球战争。联合 国正是在这场浩劫中诞生，成为人类第二次努力将国际法置于国际无政府状态之上的重要成果。正如《宪章》所言，联合国的成立是为了 "欲免后世再遭今 代人类两度身历惨不堪言之战祸"。 鉴于我们已身处核时代，绝不能重蹈覆辙 —— 否则人类将走向毁灭，再也没有第三次机会履行这一使命。 演讲要点： 自1947年以来，美国外交政策多次通过武力、秘密行动和政治操纵，在其他国家推动政权更迭。这是有充分历史记录佐证的事实：仅1947年至 1989年间，美国就尝试了70次政权更迭行动。 在委内瑞拉变局发生后，全球知名经济学家、国际事务资深学者杰弗里·萨克斯在线在 ...

据媒体8日报道，京东集团正考虑首次发行点心债券，潜在募资规模约100亿元人民币。这批离岸人民币 计价债券期限将在10年内。京东的考虑仍处于初步阶段且可能改变，包括发行时间在内的细节尚未确 定。 京东发言人回应媒体时表示，公司现阶段没有发行此类债券的计划，且"不对市场传言或猜测发表评 论"。 随着中国外卖行业价格战愈演愈烈，阿里巴巴集团和美团等电商企业纷纷涌入点心债市场，为业务扩张 融资。许多中国科技公司也转向离岸债券市场寻求资金，以支持其向人工智能领域的转型。 去年10月，美团发行了约70亿元人民币的点心债券。在此之前，腾讯控股和百度也进行了类似发行，两 家公司去年合计出售了234亿元人民币离岸债券，创下该行业历史最高纪录。阿里巴巴则在2024年底发 行了170亿元人民币离岸债券。 在AI生成视频领域取得进展的TikTok同行快手科技，也在考虑首次离岸债券发行，目前正与投资银行合 作筹备潜在交易。 风险提示及免责条款 市场有风险，投资需谨慎。本文不构成个人投资建议，也未考虑到个别用户特殊的投资目标、财务状况或需要。用户应考虑本文中的任何 意见、观点或结论是否符合其特定状况。据此投资，责任自负。 点心债券因融资 ...

Core Viewpoint - Major Wall Street investment banks are indicating that the supercycle for AI hardware has just begun, with TSMC positioned as an essential "toll booth" in this growth phase [1] Group 1: Target Price Adjustments - JPMorgan raised TSMC's target price to NT$2,100, while Nomura set a higher target of NT$2,135, reflecting strong demand and limited supply [1][6] - Both banks maintain a "overweight" or "buy" rating on TSMC, suggesting significant upside potential in the current stock price [6] Group 2: Revenue Growth Projections - JPMorgan forecasts a 30% increase in TSMC's dollar revenue by 2026, driven primarily by AI demand from data centers, with a continued growth rate of over 20% in 2027 [1][4] - Nomura emphasizes that TSMC's strong pricing power is a result of severe supply shortages, which will support revenue growth [1] Group 3: Advanced Process Technology - TSMC is expected to maintain over 95% market share in the AI accelerator market at the N2 node, with no signs of losing market share to competitors like Intel and Samsung [2] - The N2 process is projected to generate revenue of $8 billion in 2026 and surge to $36 billion by 2027, indicating rapid growth in this segment [4] Group 4: Capital Expenditure and Profit Margins - TSMC's capital expenditure is expected to approach $50 billion in 2026 and potentially exceed $55 billion in 2027 to meet rising demand [4][8] - Gross margins are anticipated to stabilize above 60%, with potential spikes to 62-63% in the first half of 2026 due to urgent orders [8] Group 5: AI Revenue Contribution - The compound annual growth rate (CAGR) for data center AI revenue has been revised upward to 57%, with expectations that AI-related revenue will account for over 40% of TSMC's total revenue by 2029 [4][8] - The transition of nearly all AI accelerators to the 3nm process by 2026 is expected to further enhance TSMC's revenue streams [4]

Core Insights - The perception that digital healthcare is a false proposition is being challenged by the surge in AI-driven healthcare demand, which has been historically underestimated due to high entry barriers and low frequency of medical needs [1] - The emergence of AI healthcare solutions like Ant Group's "Afu" and OpenAI's Health signifies a shift from traditional search engine-dominated medical information to AI agents managing personal health [5][16] Group 1: Ant Group's "Afu" - Ant Group's "Afu" saw its monthly active users double to over 30 million within a month, with daily inquiries exceeding 10 million [2][6] - "Afu" integrates with major smart device brands and connects to 5,000 hospitals and 300,000 doctors, providing a comprehensive healthcare service from monitoring to consultations [8] - A significant 55% of "Afu" users come from third-tier cities, highlighting the demand for accessible healthcare solutions in underserved markets [8] Group 2: OpenAI's Health - OpenAI launched ChatGPT Health on January 7, 2026, creating a separate entry point for health-related inquiries to address privacy concerns [3][9] - The architecture of ChatGPT Health ensures that health-related data is stored separately, preventing any cross-contamination with other data, thus fostering user trust [10] - OpenAI partnered with b.well Connected Health to address the challenge of fragmented medical data, allowing users to access their medical records seamlessly [12][13] Group 3: AI Healthcare Functionality - ChatGPT Health aims to manage the entire lifecycle of user health by integrating static medical records and dynamic physiological data [14] - The service can generate personalized health recommendations and actionable plans, demonstrating a significant advancement over traditional healthcare models [15] - Both "Afu" and ChatGPT Health are evolving to provide end-to-end healthcare services, reflecting a common demand for AI-driven healthcare solutions in both Eastern and Western markets [15] Group 4: Valuation Logic in AI Healthcare - The competitive landscape of healthcare is shifting from search engine-based information retrieval to conversational AI services, diminishing the value of traditional health information platforms [17] - The role of offline service providers is transforming as they become data asset managers rather than just service providers, emphasizing the value of high-quality real-world data [18] - Trust is becoming a critical currency in AI healthcare, with privacy assurances and data handling practices being central to user engagement and future monetization strategies [19]

大规模核电建设的历史记录让投资者望而却步。据T. Rowe Price发布的一份报告，美国要在2050年实现 产能翻两番，从2030年起每年需新增15吉瓦（GW）的装机量，这一速度将超过1974年创下的历史峰 值。该报告强调，"成本和可融资性"是扩张计划的最大威胁。 作为前车之鉴，佐治亚州的Vogtle电站3号和4号机组在2023年和2024年才投入运营，不仅延期七年，预 算更超支180亿美元。其建设成本飙升至每千瓦15000美元，约是韩国核电项目的五倍，也显著高于印 度。 特朗普政府正试图通过激进的核能扩张计划，确保美国在人工智能（AI）竞赛中占据领先地位。然 而，尽管白宫提出了到2050年将全美核电产能翻两番的雄心壮志，但面临高昂的建设成本、监管挑战以 及私营部门对财务风险的担忧，市场专家与业内人士对这一愿景的现实性表达了广泛质疑。 最新动态显示，特朗普政府已将核能视为解决AI数据中心用电缺口的关键。政府在2024年10月与私募 股权巨头Brookfield及反应堆设计商Westinghouse签署了价值800亿美元的合作伙伴关系，计划建设8座大 型核电站。同时，政府正通过提供数十亿美元贷款，推动包括密歇 ...

随着AI大周期的开启，全球科技产业正处于结构性变革的初期。尽管市场对资本开支与回报节奏仍有 讨论，但AI对生产力的潜在提升已确立了其长周期增长的主旨。2026年将成为观察AI基础设施约束、 芯片竞争格局演变及终端应用整合的关键窗口期。 据追风交易台消息，汇丰Nicolas Cote-Colisson分析团队7日发布报告，认为AI目前正处于大周期的初始 阶段。尽管2025年10月曾出现资本开支预期领先于初期营收的情况，但考虑到AI发展的速度及其对全 球超过110万亿美元GDP的潜在带动作用，市场对2026年的前景保持乐观。分析认为，风险、机遇与叙 事重点将集中在六个核心维度。 在宏观层面，汇丰对2026年的美股表现持乐观态度，将S&P 500的目标价定为7500点。分析师建议投资 者将AI贸易从单纯的基建供应商扩展至应用者和推动者。 云端产能受限与资本开支上行 全球三大云巨头（亚马逊、微软、Alphabet）订单积压严重，但产能扩张受限于数据中心建设周期。汇 丰预计产能紧张状况将持续至2026年。 在此背景下，科技巨头的投入意愿依然强劲。Alphabet、Meta和亚马逊均暗示2026年资本开支将出 现"显著/ ...

贵金属市场正经历一场由指数再平衡引发的流动性冲击。 1月8日亚市尾盘，贵金属市场再度集体调整，现货黄金短线跳水，最低跌至4415美元附近。COMEX期银跌超2%，现货白银跌超 3%，最低触及75.58美元。此外，铂金和钯金、铜等金属同样下跌。 （现货黄金30分钟走势图） （现货白银30分钟走势图） （纽约铂金30分钟走势图） （纽约钯金30分钟走势图） （LME期铜30分钟走势图） 贵金属市场的本轮调整直接关联彭博大宗商品指数(BCOM)的年度权重再平衡，该机制已于1月8日盘后启动，将持续至14日。此次 调整将黄金权重从20.4%下调至14.9%，白银权重从9.6%大幅削减至3.94%，迫使追踪该指数的被动基金进行机械性仓位调整。 | Figure 1: Target and current weights for BCOM index | | --- | | | | | | | Rebalancing demand (supply) | | --- | --- | --- | --- | --- | --- | | Group | Commodity | Ticker | 2026 Target Weigh ...

Core Insights - The discussion emphasizes that the current AI revolution is unprecedented in scale, comparable to electricity and microprocessors, and is still in its early stages [4][6][12] - The rapid decline in AI costs, described as "hyper deflation," is expected to drive explosive demand growth [4][6][35] - Historical patterns suggest that shortages in AI infrastructure, such as GPUs, will lead to oversupply and further cost reductions in the future [4][6][35] AI Technology Landscape - AI is viewed as a transformative technology that surpasses the internet in magnitude, with its current phase being very early [6][12] - The cost of AI is decreasing at a rate faster than Moore's Law, leading to significant demand growth [6][35] - The future AI market structure will resemble the computer industry, with a few "god-tier models" at the top and numerous low-cost "small models" at the periphery [6][10] US-China Competition - The emergence of Chinese models like DeepSeek and Kimi has surprised both Washington and Silicon Valley, indicating significant progress in open-source models from China [5][7][52] - The competition is characterized as a dual hegemony, with both the US and China being the primary players in AI development [5][7][51] Business Models and Pricing - A shift from "pay-per-token" to "value-based pricing" is occurring in AI applications, allowing companies to capture more value from productivity gains [9][34] - AI startups are moving beyond being mere wrappers around large models and are integrating their own models, enhancing their competitive edge [9][34] AI Democratization - Advanced AI technologies are becoming accessible to the public, breaking down barriers and allowing widespread use [6][21] - Despite public fears about AI replacing jobs, actual behavior shows a rapid adoption of AI technologies [6][21] Regulatory Environment - The regulatory landscape is shifting, with a reduced risk of federal-level restrictions on AI, as there is bipartisan support for innovation [7][59] - State-level regulations are emerging, leading to a fragmented legal environment that could hinder progress [59][60] AI Chip Development - The AI chip market is expected to see significant investment and innovation, with a potential oversupply in the coming years as companies ramp up production [4][45] - The competition in chip development is intensifying, with both established companies and startups entering the market [45][49]

Core Insights - Xiaohongshu has initiated a significant product internal test for a "paid notes" feature, introducing three payment models: high-definition original image downloads, single note paid reading, and paid unlocking of note collections [1][3] - The platform's user base is substantial, with over 339 million monthly active users (MAU) as of 2025, and a high engagement level, which provides a solid foundation for exploring paid content [3][4] Group 1: Paid Content Features - The three payment models include: 1) HD original image downloads for visual creators, 2) single note paid reading allowing users to unlock full content after a free preview, and 3) paid unlocking of entire series for serialized content [1][3] - The internal test for the paid notes feature is invitation-based, with relatively lenient entry requirements for creators [1] Group 2: Market Position and User Engagement - Xiaohongshu's user demographic is predominantly young, with 85% of users born after 1995, and 66% from first-tier and new first-tier cities, indicating a high potential for monetization [3][4] - The platform has previously launched initiatives like the "Xiaohong Card" for e-commerce, which was later suspended due to insufficient preparation, highlighting the challenges in its commercialization efforts [2] Group 3: Challenges and Considerations - The introduction of paid content may disrupt the community's established norm of free information access, necessitating a balance between community atmosphere and monetization [6][7] - Users' willingness to pay for content is uncertain, as they are accustomed to free access, and developing a culture of paid content may require time and high-quality offerings [6][7] - The platform must enhance copyright protection mechanisms to encourage creators and ensure a sustainable content ecosystem [6][7]