USA | Internet Equity Research June 24, 2025 Cannes 2025: Creative Disruption in the AI Era We tuned in remotely to the annual Cannes Lions Festival of Creativity, which brings together the top advertising & marketing leaders. We left encouraged by AI's ability to innovate ad formats & performance by streamlining creative production and campaign execution. We are bullish on META, GOOGL, & AMZN's position to benefit as advertisers & marketers embrace AI to reach broader audiences across more formats with the ...

Global Research Highlights Halftime Report & All Things "AI" Investment Strategy Resilient growth in a precarious world As we close out the first half of the year, BofA's Head of Global Econ, Claudio Irigoyen, upgrades our 2025 global growth forecast by 20bp, largely explained by China benefiting from the trade truce. We expect the global economy to grow 3% in 2025 and 2026, accelerating to 3.3% in 2027. We forecast global inflation to hover around 2.5% as economies absorb the tariffs shock. While trade de- ...

摘要 2025 年谷歌对 800G 光模块的需求预计约为 350 万支，季度需求环比 增长 20%-30%，同比增长 50%-60%。2026 年，谷歌在 AI 领域的投 入预计将带来约 50%的需求增长，采购量约为 400 万支 800G 光模块 和 100 万支 1.6T 光模块。 Lumina 收购后，公司与谷歌的合作关系得以维持，且在北美市场份额 呈现增长趋势，并拓展了 Oracle 和 Vivo 等新客户，推动产能扩张。预 计到 2026 年，公司产能将至少翻两到三倍。 1.6T 光模块行业需求推迟约半年，预计 2025 年第三季度开始小批量出 货，2026 年实现批量出货。公司计划 2025 年第四季度开始小批量出 货 1.6T 光模块，可靠性验证已基本完成，预计 2025 年出货量最多约 5 万支。 微软预计 2025 年第三季度开始采购 800G 光模块，2026 年泰国工厂 的季度产能有望达到 15 万支，主要供应微软，预计占据微软总份额的 约 20%。 Q&A 2025 年和 2026 年全球 800G 光模块的整体需求量预计将如何变化，主要驱 动因素是什么？ 光模块行业专家交流 2026 ...

华泰证券｜AIDC 产业更新 20260629 摘要 海外算力基建恢复正轨，英伟达和 Corewave 业绩向好，分别侧重训练 和推理需求，博通和 Marvell 的 ASIC 收入预期乐观，甲骨文 OCI 业务 显示推理需求增长，预示全球 AI 进入闭环。 国内 AIDC 产业链受芯片制约、模型突破和应用落地等多重因素影响， H20 禁令影响有限，头部数据中心上架率稳定，液冷、光模块、交换机 等配套环节业绩爆发，光迅、华工、锐捷网络、星网锐捷等企业受益。 国内互联网大厂如阿里、字节跳动已宣布大规模投资 AI，尽管短期受贸 易战影响，但长期来看，国内 AI 问题有望通过芯片突破（寒武纪、华为 Cloud Matrix 384）、模型提升和应用市场优势解决。 建议关注具有核心龙头客户订单且上架率稳步提升的国内 AIDC 龙头标 的，如润泽科技、奥飞数据港、世纪互联和万国数据，这些公司估值较 低，未来有望显著改善。 英伟达 GPU 主要面向训练需求，Marvell 的 ASIC 收入预期增长，甲骨 文 OCI 业务不依赖少数大客户，收入来自各行业推理和应用需求，反映 海外产业链对未来发展的信心。 Q&A 海外和国 ...

Group 1: Pershing Square Capital Management's Portfolio Overview - A significant portion of Pershing Square Capital Management's portfolio, specifically 41.6%, is invested in three companies: Alphabet, Uber Technologies, and Chipotle Mexican Grill [2] - Alphabet constitutes approximately 14% of the portfolio, with over 5.7% in class A shares and nearly 8.3% in class C shares [4][5] - Uber Technologies represents 18.5% of the portfolio, being the largest holding as of the first quarter [8] - Chipotle Mexican Grill accounts for about 9.1% of the portfolio [12] Group 2: Alphabet Insights - Alphabet's stock underperformed in the first half of the year despite strong financial results, likely due to market concerns over regulatory risks, particularly an antitrust lawsuit [4][5] - The company is well-positioned for future growth in cloud computing and artificial intelligence, which are still in early stages [6] - Alphabet's YouTube platform and its competitive advantages, such as network effects and switching costs, contribute to its attractiveness for long-term investors [7] Group 3: Uber Technologies Insights - Uber has transformed into a profitable company, with first-quarter revenue growing 14% year over year to $11.5 billion and net income reaching $1.8 billion [9] - The company's competitive edge is reinforced by its network effect, as more drivers enhance its attractiveness to clients, significantly outpacing competitor Lyft in trips and gross bookings [10] - Long-term prospects for Uber are promising, especially as younger generations are driving less, increasing demand for ride-hailing services [11] Group 4: Chipotle Mexican Grill Insights - Chipotle's stock has faced challenges in 2025 due to potential tariff impacts and weak foot traffic, but it remains a consistently profitable business with strong margins [12][13] - The company is actively expanding, having opened 57 new locations in the first quarter, with a long-term goal of reaching 7,000 locations in the U.S. and Canada [14] - Despite current struggles, the recent dip in stock price presents a buying opportunity for long-term investors [14]

Investment Rating - The report maintains an "Accumulate" rating for the industry [5] Core Insights - The demand for pre-training and inference remains strong, with significant updates expected for top models like GPT-5 and Grok 4, which are anticipated to enhance capabilities in natural language processing and coding [1][11] - The rapid advancement in coding applications is evident, with companies like Cursor achieving a valuation of $9.9 billion and an annual recurring revenue (ARR) exceeding $500 million, reflecting a 60% increase in just two months [2][19] - The launch of Ant Group's AI health application AQ marks a significant step in the application of AI in healthcare, providing over 100 services and connecting with numerous hospitals and doctors [3][28] Summary by Sections Pre-training and Inference Demand - OpenAI's CEO announced the potential release of GPT-5, which will integrate features from previous models to create a more powerful system [1][11] - Grok 4 is set to be released soon, focusing on programming capabilities [1][11] - The training data and computational power for models like Qwen3 have significantly increased, with Qwen3 utilizing approximately 36 trillion tokens, nearly double that of its predecessor [1][13][14] Coding Applications and Market Growth - Cursor's rapid commercialization has led to a valuation increase to $9.9 billion, with an ARR surpassing $500 million, indicating a strong market demand for AI programming tools [2][19] - The adoption rate of AI in U.S. enterprises is projected to reach 9.2% by May 2025, up from 7.4% in Q4 2024 [2][20] AI in Healthcare - Ant Group's AQ application offers extensive health management services and has been recognized for its advanced AI capabilities, including multi-modal interactions [3][28][29] - The application has been developed using over one trillion tokens of medical data, achieving high standards in safety and privacy assessments [3][29] Investment Opportunities - The report suggests focusing on companies involved in computing power, including Cambricon, Alibaba, and others, as they are expected to benefit from the growing demand for AI applications [4][31]

Group 1 - OpenAI has begun renting Google's TPU chips for the first time on a large scale, reducing its reliance on NVIDIA's GPUs and alleviating pressure on Microsoft's data centers [1][2] - OpenAI's demand for computing power has surged, with paid subscribers for ChatGPT increasing from 15 million at the beginning of the year to over 25 million, alongside hundreds of millions of free users [1] - Companies like Amazon, Microsoft, OpenAI, and Meta are developing their own inference chips to decrease dependence on NVIDIA and lower long-term costs [1][2] Group 2 - OpenAI spent over $4 billion on NVIDIA server chips last year, with training and inference costs each accounting for half, and is projected to spend nearly $14 billion on AI chip servers by 2025 [2] - The shift to Google's TPU was driven by the explosive popularity of ChatGPT's image generation tool, which increased pressure on OpenAI's inference servers [2] - Google has been developing TPU chips for about a decade and has provided this service to cloud customers since 2017, with other companies like Apple and Cohere also renting Google's TPU [2][4] Group 3 - Meta is also considering using TPU chips, indicating a broader trend among major AI chip customers [3] - Google Cloud continues to rent out NVIDIA-supported servers, as they remain the industry standard, generating more revenue than renting TPUs [4] - Google has ordered over $10 billion worth of the latest Blackwell server chips from NVIDIA, starting to provide them to select customers since February [4]

图片来源： Unsplash 全球最大的英伟达人工智能芯片客户之一 OpenAI ，近期开始租用谷歌的 AI 芯片来驱动 ChatGPT 等 产品， 据知情人士透露，这是该公司首次大规模采用非英伟达芯片。 • OpenAI 希望谷歌 TPU 能帮助降低其运营成本，此举是 OpenAI 为减少对微软和英伟达依赖的多元 化努力之一 • 谷歌以"硬件+云绑定"策略（TPU仅限云租赁）争夺市场份额 这一举措反映出 OpenAI 正逐步减少对微软数据中心的依赖，同时可能推动谷歌的张量处理单元（ TPU ）作为更廉价替代方案，挑战主导 AI 芯片市场的英伟达图形处理单元（ GPU ）。 该协议也表明，谷歌长期在几乎所有与 AI 相关的软硬件领域开发技术或业务的战略可能正在奏效。 这位知情人士表示， OpenAI 希望通过租用谷歌云的 TPU 来降低推理计算成本——该术语指在 AI 完 全开发完成后在服务器上运行的过程。 OpenAI 的计算需求正快速增长： ChatGPT 的付费订阅用户可能已超过 2500 万，较年初的 1500 万 大幅增长，每周还有数亿用户免费使用该服务。 OpenAI 主要通过微软和甲骨文租用英伟 ...

以下文章来源于AGI接口 ，作者余杨 AGI接口 . 出品｜虎嗅科技组 6月28日消息，在谷歌的最新播客中，机器人在Gemini人工智能技术的加持下，未经教授就完成了 一次"扣篮"动作，展现出迅速适应新场景的能力。 两天前的洛杉矶，运行着Gemini Robotics On-Device 模型的机器人还在RSS2025（机器人软件与 系统大会）上完成了全球首个互动式现场展示。硅谷AI观察人士Ted Xiao表示，"它仿佛是对未来 的神奇一瞥——只需与你的机器人对话，它就会做出适当的回应，并尝试执行一些合理的操作。 新的物体、新的命令、新的技能。" 作者｜余杨 编辑｜苗正卿 头图｜视觉中国 AI卷起的财富风暴。 谷歌Gemini Robotics On-Device 模型发布于6月24日，AI 被引入了机器人，在整个过程中不需要 持续的互联网连接，机器人实现了脱机工作。 谷歌在X上表示，这是自己首个"视觉-语言-动作"模型，旨在能够帮助机器人更快、更高效地适应 新任务和新环境。如果我们稍加回忆，今年2月底，Figure发布的"历史上最重大的人工智能更 新"——Helix也是一款用于通用人形机器人控制的"视觉-语 ...

Core Viewpoint - OpenAI has begun renting Google's TPU chips for the first time on a large scale, marking a significant shift away from its reliance on NVIDIA chips, which have dominated the AI chip market [2][4]. Group 1: OpenAI's Shift to Google TPU - OpenAI's decision to use Google TPU chips is driven by a surge in demand for computational power, with paid ChatGPT subscriptions increasing from 15 million to over 25 million since the beginning of the year [2][3]. - The collaboration with Google allows OpenAI to reduce its dependency on Microsoft's data centers and presents an opportunity for Google to challenge NVIDIA's GPU market dominance [2][4]. - OpenAI's spending on NVIDIA server chips exceeded $4 billion last year, with projections indicating that spending on AI chip services could approach $14 billion by 2025 [3]. Group 2: Industry Trends and Competitors - The increasing demand for AI capabilities has prompted several companies, including Amazon, Microsoft, OpenAI, and Meta, to develop their own inference chips to lessen reliance on NVIDIA [2][4]. - Meta is also considering the use of TPU chips, indicating a broader trend among major AI players to explore alternatives to NVIDIA [5]. - Google has been developing TPU chips for about a decade and has been offering this service to cloud customers since 2017, with other companies like Apple and Cohere also utilizing Google Cloud's TPU [4][5].