Key Points 云栖大会上，吴泳铭提出ASI（超级人工智能）的技术发展的三个阶段，分别是涌现智能、自主行动和自主学习； 目前我们处于第二阶段，第三阶段需要模型能够自己为自己的升级迭代，不仅学习人类归纳的知识，还要自己学习归纳出新知识； 未来的AI时代中，家庭、工厂、公司，都会有众多Agent和机器人，甚至未来每个人可能「都需要使用100张GPU芯片为我们工作」； 在他的框架中，阿里云将成为AI时代的计算机，千问模型是跑在这台超级计算机上的操作系统； 为此，阿里在3年3800亿元AI基础设施预算外，还要追加投入； 「开源+全栈研发能力」目前仍是阿里云的护城河，但如果仅以token消耗为维度来计算，火山引擎已超过阿里云。 在资本市场，有想法往往比有结果值钱。 阿里巴巴已经两次证明了这一点。一次是8月29日，它发布了2025年第二季度财报（2026财年第一财季，截至2025年6月30日），阿里中国电商集团经调整 EBITA同比下降14%，经营利润同比下降3%，且新增饿了么、淘宝闪购、飞猪等业务，但公司股价不跌反涨——拐点是财报发布后的电话会上，阿里巴 巴集团CEO吴泳铭和阿里中国电商集团CEO蒋凡给出了阿里巴 ...

This article has been reviewed according to Science X's editorial process and policies . Editors have highlighted the following attributes while ensuring the content's credibility: Google CEO Sundar Pichai walks to lunch at the Allen & Company Sun Valley Conference on July 9, 2025. Caught off guard by ChatGPT and mocked for early blunders with its own generative artificial intelligence efforts, Google has pulled off a dramatic turnaround in just one year, becoming a major player in consumer-facing AI. "T ...

Core Insights - The article provides an in-depth analysis of Google's Optical Circuit Switch (OCS) technology, its components, and its implications for the industry, highlighting the potential for improved efficiency and reduced latency in data transmission [1] Group 1: Google's AI Momentum - Google's AI performance has been impressive, with the launch of Gemini 2.5 Flash Image leading to 23 million new users and over 500 million images generated within a month [2] - The company has released several multimodal model updates, showcasing its leadership in AI research and development [2] Group 2: OCS Technology Overview - OCS technology aims to eliminate multiple optical-electrical conversions in traditional networks, significantly enhancing efficiency and reducing latency [5][6] - The article discusses the differences between OCS and traditional electrical switches, emphasizing OCS's advantages in low latency and power consumption [14][16] Group 3: OCS Technical Solutions - The main OCS technologies include MEMS, DRC, and piezoelectric ceramic solutions, with MEMS being the dominant technology, accounting for over 70% of the market [10][12] - MEMS technology utilizes micro-mirrors to dynamically adjust light signal paths, while DRC offers lower power requirements and longer lifespan but slower switching speeds [10][12] Group 4: Performance and Application Differences - OCS is more suitable for stable traffic patterns where data paths do not need frequent adjustments, while traditional electrical switches excel in dynamic environments [14][30] - OCS can achieve approximately 30% cost savings over time due to its longevity and lower energy consumption, despite higher initial costs [16] Group 5: Key Components of OCS - The article details critical components of OCS, including laser injection modules and camera modules for real-time calibration, ensuring long-term stability [19][20] - Micro-lens arrays (MLA) are essential for stabilizing light signals, with increasing demand expected as OCS deployment grows [26][27] Group 6: CPO vs. OCS - CPO technology integrates switching chips and optical modules to reduce latency and power consumption, making it suitable for rapidly changing data flows [29][30] - OCS, on the other hand, is ideal for scenarios with predictable data flows, such as deep learning model training, where low latency and power efficiency are critical [30] Group 7: Google's OCS Implementation - Google employs a "self-design + outsourcing" model for its MEMS chips, ensuring compatibility with its OCS systems and optimizing performance parameters [31]

Fun new features in @YouTube Shorts: Veo 3 will generate video clips with integrated audio from a single text prompt, and Lyria 2 powers ‘Speech to song’ which can turn video dialogue into a soundtrack!Google DeepMind (@GoogleDeepMind):Your next viral video could start with a single prompt thanks to AI. 📹A custom version of our Veo 3 Fast model is now available in @YouTube Shorts, generating clips with sound. Rolling out in 🇺🇲🇨🇦🇬🇧🇦🇺🇳🇿#MadeOnYouTube https://t.co/LY0h8YkqT6 ...

Come along. Let's fly away. [Music]. ...

A free simplified version of Veo 3 is available to app users to make quick vertical videos using just a text prompt. ...

Core Insights - Alphabet has reached a market valuation of US$3 trillion, becoming the fourth company to achieve this milestone, joining Nvidia, Microsoft, and Apple [1] Group 1: Infrastructure Advantage - Alphabet operates 33 submarine cables spanning over two million miles, which supports its vast data needs and enhances its internet infrastructure [2] - The company is one of the largest manufacturers of data centers, allowing it to maintain low costs and offer free software, a significant advantage over competitors [3] - This infrastructure is crucial for all of Alphabet's operations, emphasizing its importance in the company's business model [4] Group 2: User Base and Product Reach - Alphabet has seven products, including Android and YouTube, each with over two billion users, showcasing its unmatched product breadth [5] - Additionally, eight other products have over 500 million users, indicating Alphabet's digital ubiquity in the market [6] Group 3: AI Developments - Alphabet has made a significant comeback in the AI sector with its Gemini platform, which has surpassed ChatGPT in iOS app downloads [7] - AI Overviews now reach over two billion monthly users, contributing to a 10% increase in global queries [8] - Gemini's latest models have attracted nine million developers, indicating strong growth potential [8] Group 4: Revenue Growth - The combined revenue run rate for Google Cloud and YouTube is US$110 billion, with Google Cloud generating US$49 billion in the past year [10] - YouTube has become the leading streaming platform in the U.S., capturing 12.8% of total TV viewing as of June 2025 [10] - The subscription business has surpassed 270 million paid subscriptions, driven by YouTube and Google One [11] Group 5: Long-term Strategy - The AI landscape is still evolving, and Alphabet's infrastructure and long-term strategy position it well for future developments [12] - The company emphasizes that success in tech is not about being first but about enduring over time, highlighting the importance of patience for investors [14]

Group 1 - Microsoft will integrate Anthropic AI technology into Office 365, ending its exclusive reliance on OpenAI for new features in applications like Word, Excel, Outlook, and PowerPoint [2] - OpenAI is also working to reduce its dependence on Microsoft by launching a recruitment platform to compete with LinkedIn [2] - The UAE has introduced a low-cost AI inference model, K2 Think, which reportedly outperforms larger models with only 32 billion parameters, based on Alibaba's open-source Qwen 2.5 model [2] Group 2 - Google has updated its Veo 3 AI video generation tool to support 1080P resolution and vertical video formats, making it more suitable for mobile devices and social media [2] - Tencent has open-sourced its mixed Yuan image model 2.1, which supports native 2K images and bilingual input, enhancing the model's ability to generate complex prompts and accurate representations [4]

Product Update - Google Photos' photo-to-video feature is being upgraded with Google's latest video-generation model, Veo 3 [1]

Group 1 - The core viewpoint of the article is that Google has rapidly transformed its position in the AI landscape, moving from a perceived "follower" to a leader through the launch of powerful products like Gemini 2.5 Pro and advancements in multimodal AI capabilities [5][8][28]. Group 2 - The launch of Gemini 2.5 Pro marked a significant turning point for Google, achieving top rankings on LMSys Chatbot Arena and demonstrating superior capabilities in text, visual, and web development tasks [13][16][19]. - Gemini 2.5 Pro scored 35 out of 42 points in the International Mathematical Olympiad (IMO), showcasing its advanced reasoning abilities and surpassing competitors like Grok 4 and OpenAI [21][25]. - The Gemini series has been consistently upgraded, dispelling doubts about Google's AI capabilities and re-establishing its position among the top-tier models in the industry [17][18][19]. Group 3 - In the multimodal domain, Google has shown a strong lead with its Gemini models, which can seamlessly process text, code, images, audio, and video [30]. - The introduction of Gemini 2.5 Flash Image (Nano Banana) has significantly enhanced image editing capabilities, allowing for complex modifications based on natural language inputs [41][43]. - Veo 3, Google's video generation model, has set new standards in the industry by achieving high fidelity in video and audio synchronization, marking a shift in AI video generation from mere dynamic images to coherent storytelling [47][51]. Group 4 - Genie 3, a general-purpose world model, allows for the creation of interactive 3D virtual environments, which could revolutionize AI training and applications in various fields, including gaming and autonomous driving [56][62][67]. - The restructuring of Google's AI teams, merging Google Brain and DeepMind, has streamlined efforts and focused resources on accelerating AI product development [69][73]. - Google Labs has been revitalized as a key driver of innovation, encouraging teams to explore and develop new AI projects rapidly [74][76][82]. Group 5 - Google is shifting its focus from purely academic research to enhancing commercial competitiveness, ensuring that innovations are not leaked to competitors [84][86]. - The company is prioritizing AI across all its core product lines, integrating AI capabilities into search, advertising, cloud services, and more, fostering a collaborative environment [89][90]. - The article concludes that Google is poised for a significant resurgence in the AI space, leveraging its extensive technological depth and breadth to reclaim its leadership position [92][94][95].