图 点击 上方 硬AI 关注我们 拼图逐渐清晰，生态才是真正杀招？ 硬·AI 作者 | Kozmon 编辑 | 硬 AI 周末，躺在公园百无聊赖刷手机的我，差点被一条消息惊掉下巴： 全球AI圈公认的权威视频生成评测榜单VBench刚刚更新了最新一期图生视频（I2V）排名，排在第一，不是大名鼎鼎的OpenAI Sora，也不是 风头正劲的谷歌Imagen Video，而是百度的视频生成模型Steamer-I2V，总分更是飙到了89.38%！ 讲真，我第一眼看到的时候也是满脸问号：百度？图生视频？榜单第一？？？这是啥情况？ 再去刷了一圈AI圈子，结果发现好多KOL也都一脸蒙圈： "什么情况？VBench榜单第一怎么突然被百度承包了？" "百度PR部门是不是放假去了？这么猛的成绩连个新闻稿都不发一下？" | Model Name (clickable) | A A Sampled by | A Evaluated by | Accessibility | | --- | --- | --- | --- | | Steamer-I2V | Steamer Team | Steamer Team | 202205162 ...

Core Viewpoint - The article emphasizes the significant increase in AI investments by major technology companies, highlighting that AI has transitioned from an experimental phase to a growth-driving operational lever, with companies like Amazon and Meta recognizing AI as a core strategic focus [2][5]. Group 1: AI Investment Trends - Major tech players are increasingly investing in AI, with Amazon's CEO stating that AI will be one of the company's largest business areas in the coming decades [2]. - Alibaba's recent financial report shows a 10% year-over-year revenue growth, with its cloud intelligence business growing by 18%, indicating a shift towards AI-driven growth [5]. - Alibaba plans to invest over 380 billion yuan in capital expenditures over the next three years for cloud computing and AI infrastructure, surpassing the total investment of the past decade [5][6]. Group 2: Technological Advancements - Alibaba's self-developed Qwen model series has gained global recognition, with the latest Qwen-3 model outperforming competitors while significantly reducing costs [9][10]. - The open-source strategy adopted by Alibaba is crucial for fostering innovation and collaboration in AI development, with over 100,000 derivative models created based on the Qwen model [10][11]. Group 3: AI Application Across Industries - The demand for cloud computing resources is surging as AI applications transition from model training to inference deployment, positioning Alibaba Cloud as a key player in this trend [13]. - AI applications are expanding beyond large enterprises to include small and medium-sized businesses, with industries such as traditional manufacturing and agriculture actively exploring AI solutions [13][15]. - Strategic partnerships with leading companies in various sectors, including automotive and telecommunications, demonstrate Alibaba's commitment to integrating AI into real-world applications [14][15].

摩根士丹利称，在刚刚结束的摩根士丹利中国BEST会议上，有80%以上的投资者表示，很可能在近期增加对中国股票的 敞口。报告称，中国市场在所有被摩根士丹利研究评为"增持/持平"的新兴市场中是最低配的，以及短期关税对中国的负 面影响实际上可能小于许多其他主要经济体和市场。 硬·AI 作者 | 董 静 编辑 | 硬 AI 资金态度重大转变，外资大举买入中国股票。 图 点击 上方 硬AI 关注我们 摩根士丹利数据显示，中国在全球新兴市场投资组合中仍然是最大的低配地区，相对于指数基准（MSCI新 兴市场指数中国权重为31.3%）低配2.4个百分点。 这在所有被摩根士丹利研究评为"增持/持平"的新兴市场中是最低配的。 另外，摩根士丹利分析指出，短期关税对宏观和企业盈利的影响是负面的，但对中国的负面影响实际上可 能小于许多其他主要经济体和市场。 例如，对于2025年实际GDP增长，摩根士丹利中国经济团队已将预测从4.5%下调至4.2%（下调幅度为 6.7%），相比之下，美国从1.5%下调至0.6%（下调幅度60%），亚洲整体从4.4%下调至4.0%（下调幅度 9.1%）。 此外，对于盈利增长预测，摩根士丹利已将2025年M ...

Core Viewpoint - The article discusses the launch of Tencent's new tool, Hu n y u a nCu st om, which addresses the "character consistency" issue in AI-generated videos, allowing for seamless video creation without altering the identity of characters [1][5]. Group 1: Single Subject Video Generation - The core feature of Hu n y u a nCu st om is single subject video generation, where users can input an image and a text description to create a coherent video with the same character, regardless of changes in scene or attire [1][5]. - The tool can recognize the character in the image and maintain their identity throughout various scenarios, achieving results comparable to top proprietary models [1][5]. Group 2: Multi-Subject Video Generation - The tool also supports multi-subject video generation, allowing users to input images of multiple characters or products and create videos where all subjects retain their distinct features while following a scripted scenario [2]. Group 3: Video Editing and Audio Integration - A "video-driven" mode enables users to edit existing videos by integrating characters or objects from images, facilitating creative content creation and scene expansion [3]. - The "audio-driven" feature allows for single subject video dubbing, where users can provide an image and an audio clip to generate videos of the character speaking or singing in various contexts, enhancing applications in digital broadcasting and online education [4]. Group 4: Technological Advancements - Hu n y u a nCu st om introduces an "identity enhancement mechanism" and "multi-modal fusion module," which significantly improve character consistency and control in video generation compared to previous AI video technologies [5]. Group 5: Availability and Applications - The single subject generation feature was open-sourced on May 9, with additional functionalities like multi-subject generation and video editing expected to be released within the month [6]. - The tool is particularly useful for video creators, short video bloggers, e-commerce operators, and advertising creatives, enhancing efficiency and effectiveness in content production [7].

Core Viewpoint - OpenAI is restructuring its management by appointing Fidji Simo as the new application CEO, which is expected to enhance its growth and commercialization efforts, particularly in diversifying revenue sources [3][4][7][12]. Group 1: Management Changes - OpenAI has appointed Fidji Simo, previously CEO of Instacart, as the application CEO, overseeing multiple core departments while Sam Altman will focus on research, computing, and security [4][5]. - This management change follows the departure of former CTO Mira Murati, leading to Altman taking a more significant role in product and research organization [7]. Group 2: Fidji Simo's Background - Fidji Simo has extensive experience in e-commerce and advertising, having previously worked at Meta where she managed various products, including the Facebook app [9][10]. - Under her leadership, Facebook introduced several innovative features that significantly increased mobile advertising revenue, contributing billions to the company's income [10]. Group 3: Financial Projections and Challenges - OpenAI anticipates its revenue will reach nearly $13 billion this year, a substantial increase from approximately $4 billion last year [15]. - The company aims to monetize free users and other products, projecting revenues from these sources to grow to $25 billion by 2029, accounting for about 20% of total revenue [12]. - OpenAI faces challenges in balancing rising infrastructure costs with profitability while competing against major tech giants like Google and Elon Musk [15][16].

图 点击 上方 硬AI 关注我们 据报道，FDA正与OpenAI就在药物评估中使用AI进行会谈，旨在缩短药物审批上市时间。这一举措意味着药物审批流程 潜在的变革，但同时也引发了对数据安全、模型可靠性以及监管框架的担忧。 硬·AI 作者 | 李笑寅 FDA现有的审查流程通常需要约一年时间，但该机构已有多种机制可以加速有前景药物的审批时间线，包 括快速通道认证和突破性疗法认证。 近年来，FDA也正推进自己的AI潜在用途研究。2023年12月，该机构发布了一项研究员职位广告，旨在 开发供内部使用的大型语言模型。 今年1月，OpenAI宣布了ChatGPT Gov，这是一个符合政府法规的自托管版聊天机器人。该初创公司还 表示正在努力获得ChatGPT Enterprise的FedRAMP中级和高级认证，这将允许它处理敏感的政府数据。 编辑 | 硬 AI FDA拥抱AI：药物审批加速器还是风险敞口？ 据媒体周三报道，美国食品药品监督管理局(FDA)正与OpenAI进行会谈， 讨论在药物评估中使用人工智 能，旨在缩短药物审批上市时间。 媒体援引知情人士消息称，OpenAI的一个小团队近几周已数次与FDA会面，同时参与讨论 ...

图 点击 上方 硬AI 关注我们 黄仁勋在采访中表示，中国人工智能芯片市场规模预计在未来几年内将达到500亿美元，这使得美国公司能否进入该市场 变得至关重要。此前新华社报道，国务院副总理何立峰会见黄仁勋时指出，中国市场投资和消费潜力巨大，始终是外资企 业投资和贸易的沃土。欢迎包括英伟达公司在内的更多美资企业深耕中国市场。 硬·AI 作者 | 鲍弈龙 01 反对限制政策 呼吁释放美国AI实力 英伟达4月公布的一季度财报显示，最近一轮限制措施导致英伟达不得不为其H20产品计提55亿美元损 失。 在过去两年里，英伟达的估值大幅上涨，主要基于其在AI芯片市场几乎不受挑战的主导地位和强劲的收入 增长。然而，H20芯片的计提可能标志着公司进入了一个新阶段， 其中收入确认的时间和数量将部分受制 于外部因素 。 黄仁勋在采访中表示： 世界渴望人工智能，让我们把美国的AI推向所有人的前沿。 黄仁勋的言论揭示了美国当前科技竞争的困境： 一方面 ，美国试图通过限制先进芯片技术出口来维护自 身优势； 另一方面 ，美国科技巨头却因此失去了广阔的海外市场，这可能会对其营收和全球竞争力产生 长期的负面影响。 编辑 | 硬 AI 英伟达CE ...

图 点击 上方 硬AI 关注我们 OpenAI计划大幅削减给微软的收入分成比例，从目前的20%降至2030年的10%，微软则希望在OpenAI重组后继续使用 后者的知识产权，双方还在就Azure云服务器协议等方面进行协商。分析指出，当前一个悬而未决的问题是：当前协议是 否涵盖OpenAI尚未发布或尚未产生收入的产品。 硬·AI 作者 | 李笑寅 编辑 | 硬 AI 据报道，OpenAI和微软就现有协议的修改已争执数月 ，包括修改20%的收入分成比例，以及微软保留使 用OpenAI知识产权权利的时间。 据接近两家公司的高管向媒体透露，双方仍在就微软在重组后的营利性部门中的股份规模进行谈判。 微软 还希望获得未来继续使用OpenAI技术的保证。 OpenAI寻求削减微软"分成"，重组计划引发权力博弈？ 据媒体周二报道，在OpenAI推进公司重组的计划中，其中一个关键部分是大幅削减与微软的收入分成比 例。 具体而言，在现有协议中，OpenAI同意在2030年之前向微软分享其20%的收入。 而OpenAI向潜在和现 有投资者表示，到2030年，它将仅与包括微软在内的商业伙伴分享10%的收入，这从某些内部文件中可以 得 ...

Core Viewpoint - Super Micro Computer's fourth-quarter revenue guidance significantly underperformed analyst expectations, indicating potential obstacles in the AI infrastructure boom [4][5]. Financial Performance - For the third quarter, Super Micro reported revenue of $4.6 billion, below the expected $5.42 billion, with adjusted earnings per share of $0.31, also lower than the anticipated $0.50 [3][4]. - The net profit per share dropped to $0.17, a substantial decline from $0.66 in the same period last year [4]. - The fourth-quarter revenue guidance is set between $5.6 billion and $6.4 billion, which is far below the market expectation of $6.82 billion [5][6]. - The full-year revenue forecast for fiscal 2025 has been revised down from $23.5-25 billion to $21.8-22.6 billion [8]. Market and Economic Context - CEO Charles Liang noted that customer hesitation between current Hopper and upcoming Blackwell GPUs has led to delayed purchasing decisions [6]. - Economic uncertainty and the impact of Trump's tariffs are beginning to manifest, contributing to the company's cautious outlook [6]. Company Challenges - The past year has been tumultuous for Super Micro, with significant stock price fluctuations and scrutiny over accounting practices following a report from Hindenburg Research alleging "accounting manipulation" [10]. - The resignation of Ernst & Young as the company's auditor raised concerns about financial reporting and internal controls [10]. Future Outlook - Despite short-term challenges, Super Micro's positioning as an AI infrastructure supplier remains attractive, utilizing chips from Nvidia and AMD [12]. - Analysts express caution regarding the company's ability to regain growth momentum amid procurement cycles, competitive pressures, and macroeconomic uncertainties [15].

Core Viewpoint - Palantir's first-quarter performance significantly exceeded expectations, with a revenue of $884 million, a 39% increase from $634.3 million in the same period last year, and the company raised its full-year guidance despite facing short-term stock price pressure due to high valuation [3][4][5]. Group 1: Financial Performance - Palantir reported a first-quarter revenue of $884 million, surpassing analyst expectations of $863 million, and showing a 39% year-over-year growth from $634.3 million [5]. - Adjusted earnings per share were $0.13, in line with market expectations, while net income rose to approximately $214 million ($0.08 per share), significantly up from about $105.5 million ($0.04 per share) in the previous year [5]. - The operating profit margin for the first quarter reached 19.9%, a substantial improvement from 12.8% in the same period last year [5]. Group 2: Business Segments - The U.S. government business grew by 45% to $373 million, while U.S. commercial revenue surged by 71% to $255 million [5]. - The strong performance in the government sector was attributed to broader adoption of Palantir's tools by U.S. government departments, with a noted surge in demand for large language models and supporting software [5]. Group 3: Valuation and Market Expectations - Despite strong financial results, Palantir's high valuation poses a risk for stock price corrections, as the company’s price-to-sales ratio stands at 102.34, the highest in the S&P 500, more than double that of Texas Pacific Land [7]. - Analysts indicate that for companies with such high valuations, investors expect not just to meet but to significantly exceed performance expectations, which explains the post-earnings stock price decline despite strong results [7]. Group 4: Future Outlook - Palantir has raised its full-year revenue guidance to between $3.89 billion and $3.90 billion, up from the previous forecast of $3.74 billion to $3.76 billion [9]. - The company anticipates second-quarter revenue to be between $934 million and $938 million, with adjusted operating income expected in the range of $1.711 billion to $1.723 billion [9].