Scroller Briefings - OpenAI Completes For-Profit Transition, Secures $250 Billion Cloud Services Deal 💼
编辑
目录
单词学习：![[09d1c185-6198-43f7-a052-04bc7eaf50da.mp3]]
新闻播报：![[d49edff8-ecd2-4eeb-a813-9f48ab850108.mp3]]
OpenAI完成营利性转型，获2500亿美元云服务订单
简报：
- OpenAI正式完成公司重组，将AI实验室转型为营利性公司OpenAI Group，在非营利OpenAI基金会监管下运作
- OpenAI基金会持有OpenAI Group 26%股份并拥有董事会任命权，微软持有约27%股份（市值1350亿美元）
- 微软与OpenAI达成2500亿美元Azure云服务采购协议，OpenAI获得"云自由"，可自由选择其他云服务商部署非API类产品
- 微软对OpenAI模型知识产权的掌控权延长至2032年，API服务仍将独家运行于Azure平台
- OpenAI承诺在基础设施方面投资约1.4万亿美元（相当于30吉瓦数据中心容量），目标每年实现1万亿美元基础设施支出
- 员工股份出售交易完成后，OpenAI估值达5000亿美元，成为全球最具价值私营公司
相关链接：
- OpenAI 完成重组：从非营利到营利，AI 未来更具潜力
- 微软与OpenAI重塑联盟：2500亿美元Azure订单背后，OpenAI获"云自由"
- OpenAI 计划每年投资 1 万亿美元推动基础设施建设
OpenAI Completes For-Profit Transition, Secures $250 Billion Cloud Services Deal 💼
Brief:
- OpenAI officially completes corporate restructuring, transforming the AI lab into a for-profit entity called OpenAI Group, operating under the oversight of the non-profit OpenAI Foundation 🚀
- The OpenAI Foundation holds a 26% stake in OpenAI Group with board appointment rights, while Microsoft holds approximately 27% (valued at dJyH6kDkfC5RlM16.answerText35 billion)
- Microsoft and OpenAI reach a $250 billion Azure cloud services procurement agreement; OpenAI gains "cloud freedom," allowing it to freely choose other cloud providers for non-API products
- Microsoft's control over OpenAI's model intellectual property extends until 2032, with API services remaining exclusively on the Azure platform
- OpenAI commits to investing approximately dJyH6kDkfC5RlM16.answerText.4 trillion in infrastructure (equivalent to 30 gigawatts of data center capacity), targeting dJyH6kDkfC5RlM16.answerText trillion in annual infrastructure spending
- Following completion of employee share sales, OpenAI reaches a valuation of $500 billion, becoming the world's most valuable private company 💰
Related Links:
- OpenAI Completes Restructuring: From Non-Profit to For-Profit, AI Future Shows Greater Potential
- Microsoft and OpenAI Reshape Alliance: Behind the $250 Billion Azure Order, OpenAI Gains "Cloud Freedom"
- OpenAI Plans to Invest dJyH6kDkfC5RlM16.answerText Trillion Annually to Drive Infrastructure Development
OpenAI计划2028年推出完全自动化AI研究员
简报：
- OpenAI CEO萨姆·奥特曼在直播中透露，预计到2026年9月实现"实习级别的研究助理"，2028年前开发出完全自动化的"合法AI研究员"
- 新的AI研究员将能够自主完成大型研究项目，而非仅协助人类科研工作
- OpenAI已完成从非营利组织向公益型公司的转型，以打破资金募集限制并吸引更多投资
- 首席科学家Jakub Pachocki指出，未来十年内人工智能有望实现"超级智能"，在多个关键领域超越人类能力
- OpenAI计划未来几年投入30吉瓦数据中心计算能力，预计总财务支出达1.4万亿美元，目标是每年在基础设施上投入一万亿美元
相关链接：
OpenAI Plans to Launch Fully Automated AI Researchers by 2028 🤖
Brief:
- OpenAI CEO Sam Altman revealed during a livestream that the company expects to achieve "intern-level research assistants" by September 2026, and develop fully automated "legitimate AI researchers" before 2028 🚀
- The new AI researchers will be capable of autonomously completing large-scale research projects, rather than merely assisting human scientific work
- OpenAI has completed its transformation from a non-profit organization to a public benefit corporation, aiming to remove fundraising restrictions and attract more investment
- Chief Scientist Jakub Pachocki noted that artificial intelligence is expected to achieve "superintelligence" within the next decade, surpassing human capabilities in multiple key areas
- OpenAI plans to deploy 30 gigawatts of data center computing power in the coming years, with an estimated total financial expenditure of dJyH6kDkfC5RlM16.answerText.4 trillion, targeting an annual investment of one trillion dollars in infrastructure 💰
Related Links:
豆包AI编程推出可视化编辑器与多Agent协作系统，实现8分钟零代码生成H5页面
简报：
- 字节跳动豆包AI编程完成重大升级，首创PPT式可视化编辑界面，零代码用户可在8分钟内生成可上线的交互式H5、数据看板或活动页面
- 支持多模态输入方式，用户可通过自然语言描述、手绘草图、参考图片、Word文档或PPT上传需求，AI自动解析意图并优化提示词
- 内置所见即所得可视化编辑器，用户可像操作PowerPoint一样直接拖拽元素、调整尺寸、修改配色，系统实时生成标准前端代码并自动保存版本支持回退
- 引入多Agent协同工作流，包括检索Agent、视觉Agent、代码Agent和预览Agent四大AI角色，由主控Agent统一调度实现"需求进，产品出"的全自动交付
- 当前版本主要聚焦前端交互场景，适用于活动H5、作品集、数据可视化等，API对接、动态数据源、用户系统等后端能力正在开发中
相关链接：
Doubao AI Programming Launches Visual Editor and Multi-Agent Collaboration System, Enabling Zero-Code H5 Page Generation in 8 Minutes ✨
Brief:
- ByteDance's Doubao AI Programming has completed a major upgrade, pioneering a PPT-style visual editing interface that allows zero-code users to generate production-ready interactive H5 pages, data dashboards, or event pages within 8 minutes 🚀
- Supports multimodal input methods—users can submit requirements through natural language descriptions, hand-drawn sketches, reference images, Word documents, or PPT uploads, with AI automatically parsing intent and optimizing prompts
- Features a built-in WYSIWYG visual editor, enabling users to drag elements, adjust sizes, and modify color schemes just like operating PowerPoint, while the system generates standard front-end code in real-time and automatically saves versions with rollback support
- Introduces a multi-agent collaborative workflow, including Retrieval Agent, Visual Agent, Code Agent, and Preview Agent—four AI roles coordinated by a master Agent to achieve fully automated "requirements in, product out" delivery 💡
- The current version focuses primarily on front-end interactive scenarios, suitable for event H5 pages, portfolios, data visualization, etc., with backend capabilities such as API integration, dynamic data sources, and user systems currently under development
Related Link:
MiniMax发布Hailuo 2.3：文字生成电影级视频，动作表情物理三大突破
简报：
- MiniMax旗下Hailuo 2.3正式发布，实现"动作零卡顿、表情真到肉眼难辨、物理真实到穿模绝迹"的电影级视频生成效果
- 在动作流畅度、面部微表情、物理交互三大维度实现突破，人物动作符合真实力学，亚洲面孔细节精准，水花、布料、烟雾等物理效果遵循现实规律
- 首创双版本策略：电影级模式适合品牌宣传和短片预告，极速模式(Hailuo 2.3 Fast)生成速度提升3倍、成本降低60%，专为社交媒体设计
- 即日起开启免费试用，用户可领取每日5条免费生成额度，支持中英文双语输入，最长生成6秒高清视频并支持商用授权
- 在中文语义理解、亚洲人种面部细节、物理真实性上相比Sora、Runway占据优势，预计将推动品牌营销、短剧制作等领域AI视频内容爆发
相关链接：
MiniMax Launches Hailuo 2.3: Text-to-Cinematic Video with Breakthroughs in Motion, Expression & Physics 🎬
Brief:
- MiniMax's Hailuo 2.3 officially launches, achieving cinematic-quality video generation with "zero motion stuttering, hyper-realistic facial expressions nearly indistinguishable to the naked eye, and physics so accurate that clipping artifacts are virtually eliminated"
- Delivers breakthroughs across three key dimensions: motion fluidity, facial micro-expressions, and physical interactions. Character movements follow real-world mechanics, Asian facial details are precisely rendered, and physical effects like water splashes, fabric, and smoke adhere to realistic laws of physics
- Pioneers dual-version strategy: Cinematic mode ideal for brand promotions and film trailers, while Fast mode (Hailuo 2.3 Fast) boosts generation speed 3x and cuts costs by 60%, specifically designed for social media content 📱
- Free trial now available—users can claim 5 free daily generation credits. Supports bilingual input (Chinese/English), generates up to 6 seconds of HD video, and includes commercial use licensing
- Holds competitive advantages over Sora and Runway in Chinese semantic understanding, Asian facial detail accuracy, and physical realism. Expected to catalyze an explosion of AI video content in brand marketing, short-form drama production, and related sectors 🚀
Related Links:
- Free Trial! Hailuo 2.3 Launch! Text-to-Cinematic Video with Authentic Motion & Expression!
- Hailuo Video App Experience
英伟达发布OmniVinci模型，数据效率提升六倍超越顶尖模型
简报：
- 英伟达研究团队发布全模态理解模型OmniVinci，在多个全模态理解基准测试中超越目前最顶尖模型19.05分
- OmniVinci训练仅使用0.2万亿Token，相较于竞争对手的1.2万亿，数据效率是其六倍
- 该模型可同时理解视觉、音频和文本，采用全模态潜在空间概念融合不同感官信息
- 架构整合了图像、视频、音频和文本等异构输入，通过OmniAlignNet模块对齐视觉和音频信息
- 采用两阶段训练方法：先进行模态特定训练，再进行全模态联合训练，显著提升联合理解能力
相关链接：
NVIDIA Releases OmniVinci Model, Achieving 6x Data Efficiency Over Top-Tier Models 🚀
Brief:
- NVIDIA's research team has released OmniVinci, an omni-modal understanding model that surpasses current state-of-the-art models by 19.05 points across multiple omni-modal understanding benchmarks
- OmniVinci was trained using only 0.2 trillion tokens, compared to competitors' 1.2 trillion, achieving 6x data efficiency 📊
- The model can simultaneously understand vision, audio, and text, employing an omni-modal latent space concept to fuse information from different sensory modalities
- The architecture integrates heterogeneous inputs including images, videos, audio, and text, aligning visual and audio information through the OmniAlignNet module
- It adopts a two-stage training methodology: modality-specific training followed by omni-modal joint training, significantly enhancing joint understanding capabilities ⚡
Related Links:
富士康美国工厂将投用人形机器人生产英伟达AI服务器
简报：
- 富士康宣布将在休斯顿工厂投用人形机器人，专门用于生产英伟达AI服务器，计划于2026年第一季度投入使用
- 该工厂将成为首批使用NVIDIA Isaac GR00T模型驱动人形机器人的生产线，富士康与英伟达计划打造全球领先的AI智能示范工厂
- 富士康正与英伟达联合开发自研人形机器人，并已测试中国厂商优必选的产品
- 富士康将继续扩大得州、威斯康星州和加州的AI服务器产能以应对需求增长
- 鸿海将在11月科技日上展示最新一代人形机器人解决方案，新产品将内置大语言模型和"AI大脑"
相关链接：
Foxconn's US Factory to Deploy Humanoid Robots for NVIDIA AI Server Production 🤖
Brief:
- Foxconn has announced plans to deploy humanoid robots at its Houston facility, specifically for manufacturing NVIDIA AI servers, with operations scheduled to begin in Q1 2026 💼
- The factory will become one of the first production lines to utilize humanoid robots powered by NVIDIA's Isaac GR00T model, as Foxconn and NVIDIA aim to build a world-leading AI-smart demonstration factory
- Foxconn is jointly developing proprietary humanoid robots with NVIDIA and has already tested products from Chinese manufacturer UBTech 🏭
- The company will continue expanding AI server production capacity in Texas, Wisconsin, and California to meet growing demand
- Hon Hai will showcase its latest generation of humanoid robot solutions at its November Tech Day, with new products featuring built-in large language models and "AI brains"
Related Link:
豆包夺原生AI应用月活第一，移动端AI用户破7亿
简报：
- 2025年三季度移动端AI应用用户规模突破7亿，其中移动端原生APP、In-APP AI和手机厂商AI助手月活跃用户分别达到2.87亿、7.06亿和5.35亿
- 豆包和DeepSeek分别以1.72亿和1.45亿月活跃用户规模占据原生AI应用前两位
- AI应用插件表现强劲，在月活跃用户中占据31个位置，前三名为百度AI搜、抖音AI搜和微信AI搜
- 蚂蚁集团旗下AI健康应用AQ以83.4%的三季度复合增长率位居行业第一，排名升至中国AI原生应用第7位，成为Top10中唯一的健康类应用
- PC端应用增速放缓，超过85%的PC网页应用月活跃用户低于50万，仅24.7%实现正增长
- 百度、阿里、腾讯和抖音的大模型更新频率分别为3.8天/次、4.6天/次、6.6天/次和8.0天/次
相关链接：
Doubao Takes Top Spot in Native AI App MAU, Mobile AI Users Surpass 700 Million 📱
Brief:
- Mobile AI application user base exceeded 700 million in Q3 2025, with monthly active users reaching 287 million for native mobile apps, 706 million for in-app AI, and 535 million for smartphone manufacturer AI assistants 🚀
- Doubao and DeepSeek ranked first and second among native AI applications with 172 million and 145 million monthly active users respectively
- AI application plugins demonstrated strong performance, occupying 31 positions among monthly active users, with Baidu AI Search, Douyin AI Search, and WeChat AI Search taking the top three spots
- Ant Group's AI health application AQ achieved an 83.4% compound growth rate in Q3, ranking first in the industry and climbing to 7th place among China's native AI applications, becoming the only health-focused app in the Top 10 💡
- PC application growth slowed, with over 85% of PC web applications having monthly active users below 500,000, and only 24.7% achieving positive growth
- Large language model update frequencies for Baidu, Alibaba, Tencent, and Douyin are 3.8 days/update, 4.6 days/update, 6.6 days/update, and 8.0 days/update respectively
Related Links:
Zoom CEO袁征：AI可让工作周缩短至三四天
简报：
- Zoom创始人兼CEO袁征在TechCrunch Disrupt 2025大会上表示，AI助手有望让人们告别"五天工作制"
- Zoom正全面融入AI技术，推出"数字分身"功能，能让AI化身代替用户开口说话，袁征曾在投资者财报会议上使用自己的AI化身现场演示
- 袁征设想未来两位企业高管可让各自的数字分身先行沟通，从而节省大量时间
- AI可帮忙整理邮件、筛选重要信息并提醒优先回复，完善Zoom的在线白板和共享文档等协作功能
- 袁征预测五年后每周工作三四天就够了，当前投资重点为"AI、AI、还是 AI"
- Zoom已更改公司全名移除"Video"字样，宣布转型为"AI工作平台"
相关链接：
Zoom CEO Eric Yuan: AI Could Shorten Work Week to Three or Four Days
Brief: 📰
- Zoom founder and CEO Eric Yuan stated at TechCrunch Disrupt 2025 that AI assistants could help people move away from the "five-day work week" 🤖
- Zoom is comprehensively integrating AI technology, launching a "digital twin" feature that allows AI avatars to speak on behalf of users. Yuan once demonstrated his own AI avatar live during an investor earnings call
- Yuan envisions a future where two business executives could have their respective digital twins communicate first, thereby saving significant time
- AI can help organize emails, filter important information and remind users to prioritize responses, while enhancing Zoom's collaborative features like online whiteboards and shared documents
- Yuan predicts that in five years, working three to four days per week will be sufficient. His current investment priority is "AI, AI, and more AI" 💼
- Zoom has changed its full company name by removing the word "Video," announcing its transformation into an "AI work platform"
Related Links:
如果对你有用的话，可以打赏哦
打赏
本文作者:topwind
本文链接:
版权声明:本博客所有文章除特别声明外，均采用 BY-NC-SA 许可协议。转载请注明出处！
目录