中芯国际2025年第四季度实现营业收入178.13亿元，同比增长11.9%；归母净利润12.23亿元，同比增长23.2%，扣非净利润同比大幅增长112.4%。 全年营收673.23亿元，同比增长16.5%，毛利率提升至21.0%。公司预计2026年第一季度收入环比持平，全年收入增速目标高于行业平均，年度资 本开支计划与2025年的81亿美元基本持平。 2月10日， $中芯国际 (00981.HK)$ / $中芯国际 (688981.SH)$ 发布2025年第四季度业绩快报。报告期内，公司实现营业收入178.13亿元，同比增长 11.9%，主要受益于需求持续回暖及国产进程订单拉动，体现为晶圆销量增长、产能利用率提升及产品结构优化。 快报核心数据如下： 第四季度归母净利润12.23亿元，同比增长23.2%； 扣非归母净利润9.47亿元，同比大幅增长112.4%； 营业利润为21.14亿元，同比下降23.0%，主要受财务费用上升与投资收益减少影响。 2025年全年，公司未经审计营业收入673.23亿元，同比增长16.5%； 归母净利润50.41亿元，同比增长36.3%； 扣非归母净利润41.24亿元，同比增长55. ...

Core Insights - OpenAI has begun testing an advertising feature for ChatGPT in the U.S., marking a significant step in its commercialization efforts [1] - The initiative aims to generate advertising revenue to cover infrastructure costs while maintaining free service for users [1][2] - The move has sparked widespread controversy and backlash from users on social media [1][4] Commercialization Path - The advertising test is based on the principle of "task consistency," allowing OpenAI to sustain its free and low-cost service tiers [2] - Ads will be displayed based on user conversation topics and past interactions, similar to traditional search engines [2] - Users can opt for an ad-free experience by upgrading to a premium subscription or manage ad personalization settings [2] Privacy Commitments and User Control - OpenAI has outlined data protection measures, ensuring advertisers cannot access user chat histories or personal details [3] - Ads targeting users under 18 will be automatically blocked, and sensitive topics will not feature ads [3] - Users have control over ad preferences, including the ability to reject specific ads and manage personalized settings [3] Market Reaction and Competitive Dynamics - The introduction of ads has led to negative feedback on social media, with users expressing disappointment and concerns about the impact on ChatGPT's utility [4][5] - Competitor Anthropic has capitalized on this sentiment by mocking OpenAI's ad strategy in a Super Bowl advertisement [5] - Some users have expressed understanding, provided that ads do not interfere with the quality of generated responses [5] Financing Prospects and Product Iteration - OpenAI is accelerating product iterations alongside the ad testing, with positive reception for its new programming model, GPT-5.3-Codex [6] - The company is pursuing a massive financing round, potentially reaching $100 billion, to support its operations and prove the viability of its business model [7] - Successful implementation of the "AI + advertising" model could lead to a ripple effect in the industry, prompting other companies to accelerate their commercialization efforts [7]

德银指出，在当前全球固定收益市场中，日本国债市场正潜伏着一头巨大的"大象"，即日本国债2年期与30年期收益率曲线（JGB 2s30s） 的趋平交易。 据追风交易台，2月9日，德银策略团队的测算指出，从估值角度来看，这一利差拥有极其惊人的收窄潜力，预计可能有高达100个基点的趋 平空间。对于投资者而言，这是一个极为显著的错配机会，意味着市场当前的定价逻辑与基本面存在严重脱节。随着日本大选结束，新政 府的政策选择将成为关键催化剂。 如果市场定价回归理性，这一结构性交易将为投资者提供极具吸引力的风险回报比，即便是设置较为宽松的止损，其潜在的上行空间也足 以支撑一次雄心勃勃的"猎象"行动。 估值维度的双重错配：被低估的利率与被高估的远期 深入分析估值层面，德意志银行的研究强调了日本债市当前存在的两个核心矛盾。 首先，考虑到当前的工资动态，日本的实际政策利率无论是从绝对值角度还是跨市场比较角度来看，都处于过低的水平。这意味着短端利 率有上行的内生动力以匹配经济现实。 政治现实的硬约束：通胀已成选民头号大敌 除了纯粹的数学估值，政治环境的变迁为这一交易提供了坚实的逻辑支撑。 德银报告指出，在一个拥有巨额私营部门储蓄的国家 ...

Core Viewpoint - SMIC reported a net profit of 1.22 billion yuan in the fourth quarter, representing a year-on-year increase of 23.2% [1] Financial Performance - The net profit for the fourth quarter was 1.22 billion yuan [1] - This reflects a growth of 23.2% compared to the same period last year [1]

Core Viewpoint - Amazon is planning to establish a new content marketplace that connects publishers with AI developers, providing infrastructure for monetizing digital content in the AI era [1][2]. Group 1: Content Marketplace Development - The platform aims to create a standardized trading mechanism allowing publishers to sell content access rights to AI companies, addressing the demand for a more sustainable payment model based on usage rather than one-time buyouts [1][3]. - Amazon's initiative is a response to Microsoft's recent actions and is a crucial step in building an AI ecosystem on top of its cloud computing dominance [2]. Group 2: Integration with AWS - Amazon is attempting to integrate this content marketplace with its core AI tools, positioning it as a business tool for publishers to host content on AWS and license it to AI companies also operating on the same platform [3]. - This approach aligns with the growing demand from publishers for a payment model based on content usage frequency, which is seen as more sustainable than traditional copyright buyouts [3]. Group 3: Industry Tensions - The relationship between publishers and tech platforms is becoming increasingly strained, with publishers expressing concerns that generative AI is altering user search habits and diverting traffic away from news sites [4]. - Legal conflicts are escalating, exemplified by a lawsuit from Penske Media against Google, alleging that AI-generated summaries harm publisher revenues [4]. Group 4: Market Challenges - Despite the rapid development of the platform, there are concerns regarding actual demand from AI companies on the buyer side of the content marketplace [5][6]. - Technical challenges exist, as some AI bots disguise themselves as human users to avoid payment, complicating the enforcement of content access fees [6]. - The competition between Amazon and Microsoft in the AI data trading infrastructure is expected to intensify, with the success of this model hinging on addressing technical vulnerabilities and attracting sufficient buyers [6].

作者 | 黄昱 具身智能领域那堵长期存在的"智力高墙"，正在被一点点打破。 2月10日，阿里巴巴达摩院正式发布具身智能大脑基础模型 RynnBrain，并一次性开源了包括业界首个 30B MoE（混合专家模型）架构在内的 7 个全系列模 型。 这一动作具有一定的里程碑意义。据介绍，RynnBrain首次让机器人拥有了时空记忆和空间推理能力，同时在16项具身开源评测榜单上刷新纪录（SOTA）， 超越谷歌Gemini Robotics ER 1.5等行业顶尖模型。 这意味着，具身智能长期面临的"时空遗忘"与"物理幻觉"枷锁正被努力解开，机器人大脑有望从简单的指令接收器进化为具备深度环境理解能力的智能实 体。 长期以来，具身模型的智能水平已成为制约机器人走向通用化的重要瓶颈，尤其是泛化能力的短板极大地限制了其在复杂物理场景中的应用。 为了突破这一瓶颈，业界形成了多条技术探索路线。 据华尔街见闻了解，一类是侧重于动作输出的 VLA 模型，虽然能直接操作物理世界，但由于高质量机器数据稀缺，极难实现跨场景的泛化；另一类则是引 入具备泛化潜力的 VLM 等大脑模型，但这类模型普遍缺乏记忆能力，动态认知受限，且普遍存在物理幻 ...

作者 | 周智宇 2026年的春节前夕，中国互联网的火药味比往年更浓烈了。 2月10日，华尔街见闻从百度内部人士处获悉，百度内部已于近期悄然启动代号为"O计划"的项目。不 同于以往的部门协作，这次由搜索和云团队联合办公，意味着项目打破部门藩篱，MEG与ACG联合作 战。 另有知情人士透露，这次合作的项目和百度App相关。百度瞄准用户的日常场景，将以百度App为枢 纽，借助文心助手，调动百度内部生态及合作伙伴的服务，解决用户的实际需求。 这一系列动作释放一个强烈信号。百度意识到，AI入口的竞争不再仅仅是前端UI的竞争，而是算力成 本、推理速度与数据调度能力的综合战争。 说白了，UI只是一个对话框，谁都能做；但真正拉开代际差距的，是背后的调度逻辑与算力账本。当 用户提出一个复杂需求，系统需要瞬间调动全生态的服务。如果调度能力差，AI就只是个会聊天的花 架子，没法真正解决问题；而如果算力效率低，单次回答的成本就会贵得离谱，商业上根本跑不通。 百度启动"O计划"并非没有征兆。百度创始人李彦宏在2026年初的一次内部分享中明确定调，训练的目 的是推理。 这意味着，推理效率将决定搜索的代际竞争。如果百度不能通过搜索与云的 ...

Group 1 - The core viewpoint of the articles indicates that AstraZeneca expects a double-digit percentage growth in adjusted earnings per share by 2026, driven by strong sales in its oncology drug business, which will offset the impact of patent expirations on core diabetes drugs [1] - AstraZeneca's fourth-quarter profits and revenues met expectations, with the company forecasting mid-to-high single-digit revenue growth for the year, similar to the 8% increase in 2025, allowing room for potential upward revisions later in the year [1][2] - The company is accelerating its entry into the weight loss drug market, having recently signed a collaboration agreement with China’s CSPC Pharmaceutical Group worth up to $18.5 billion, which includes access to CSPC's long-acting peptide technology platform for developing next-generation weight loss therapies [1] Group 2 - AstraZeneca's self-developed oral GLP-1 drug, elecoglipron, has entered late-stage clinical trials, becoming a key candidate in its obesity treatment pipeline [1] - The company has confidence in managing pricing impacts and is focused on achieving its long-term revenue target of $80 billion by 2030, with several new drugs in clinical development being critical for maintaining growth momentum [2] - Under CEO Pascal Soriot's leadership, AstraZeneca has established a leading position in oncology and is the second-largest listed company on the London Stock Exchange, with key clinical data for new drugs targeting diseases like lung cancer and COPD expected to be released this year [3]

Core Viewpoint - Alphabet is launching an unprecedented financing campaign globally, having completed a record $20 billion bond issuance and is now entering the European market with Swiss franc bonds and rare 100-year pound bonds to meet its $185 billion AI infrastructure investment needs [1]. Group 1: Bond Issuance Details - Alphabet has begun selling Swiss franc-denominated bonds with maturities of 3, 6, 10, 15, and 25 years, marking its first entry into the Swiss bond market [1]. - The company is also planning to issue its first pound bonds with maturities of 3, 6, 15, 32, and 100 years, which would be the first issuance of such long-term bonds by a tech company since Motorola in 1997 [1]. - The recent bond issuance is driven by Alphabet's announcement of a record capital expenditure plan of $185 billion for this year, which is double last year's spending and exceeds the total of the past three years [1]. Group 2: Demand and Market Trends - The $20 billion dollar bond issuance attracted over $100 billion in peak subscription orders, making it one of the strongest demand corporate bond issuances in history [2]. - The strong market demand led to significant pricing tightening, with the 3-year bond pricing at a premium of only 0.27 percentage points over U.S. Treasuries, down from an initial discussion of 0.6 percentage points [2]. - The issuance trend is part of a broader financing wave in the tech sector, with the five major AI cloud computing giants issuing $121 billion in corporate bonds last year, compared to an average of $28 billion from 2020 to 2024 [3]. Group 3: Future Projections - Morgan Stanley projects that borrowing by cloud computing giants will surge from $165 billion in 2025 to $400 billion [3]. - The total investment-grade bond issuance is expected to reach a record $2.25 trillion this year, driven by the current wave of bond issuances [3]. - Bloomberg industry research estimates that capital expenditures for AI, cloud infrastructure, and data centers will total $3 trillion by 2029 [3]. Group 4: Market Concerns - The surge in bond issuance is raising concerns about bond valuation pressures, with expectations that the large volume will widen corporate bond spreads [4]. - Current market conditions are likened to those of 1997-98 or 2005, indicating potential credit performance issues, although not necessarily at the end of a cycle [4].

埃隆·马斯克（Elon Musk）正在筹谋一盘价值1.25万亿美元的大棋。 就在上周，这位科技狂人透露了一项惊人的计划：将SpaceX与他亏损的AI初创公司xAI合并。这不仅仅 是一次资本层面的运作，其核心逻辑在于一个看似科幻的商业赌注——AI的未来不在地球，而在轨道 上。 马斯克坚信，在未来三年内，由太阳能驱动、利用太空真空冷却的庞大卫星群，将成为生成AI算力最 便宜的方式。 这究竟是马斯克为了IPO讲出的又一个动听故事，还是AI产业不得不走的必经之路？ 巨头们的"星际算力"竞赛 马斯克并非唯一仰望星空的科技巨头。 虽然前xAI员工吐槽这个想法"完全未经证实"，但亚马逊创始人贝佐斯（Jeff Bezos）早已入局。他不仅 联合管理着神秘的AI初创公司"普罗米修斯计划"（Project Prometheus），更在去年10月预言未来将建 设"吉瓦级的太空数据中心"。 与此同时，谷歌也没闲着。他们计划在2027年初与Planet合作启动"Suncatcher项目"，将两颗搭载TPU芯 片的原型卫星送入低地球轨道进行"学习任务"。 Aetherflux的创始人Baiju Bhatt一针见血地指出：AI繁荣创造了" ...