July 8, 2025 09:00 PM GMT Asia Technology | Asia Pacific AI ASIC: Reconciling Trainium2 chip shipments We attribute the AWS Trainium2/2.5 chip shipments mismatch (between semis and systems) to unstable PCB yield rates. We expect chip shipments to reach around 1.1mn in 2025. Reason for this report: We conducted follow-up research on AI Supply Chain: AI ASIC dynamics: Trainium and TPU, as some investors expressed confusion about the wide range of AWS Trainium2/2.5 chip shipment assumptions, from 1mn to 2mn un ...

Group 1 - Amazon is considering a significant additional investment in AI startup Anthropic, potentially amounting to several billion dollars, to strengthen their strategic alliance [1] - This new investment would build on Amazon's previous investment of $8 billion, aiming to enhance their competitive position in the global AI commercialization landscape against Microsoft and OpenAI [1][3] - Amazon's investment will ensure it remains one of the largest shareholders in Anthropic, which has also received over $3 billion from Google [1][3] Group 2 - Amazon's recent investment in Anthropic is part of a series of investments, including $2.75 billion in March 2022 and $1.25 billion in September 2023, marking Amazon's largest external investment in nearly 30 years [3] - The collaboration includes a major data center project, Project Rainier, which will support Anthropic's computing needs and is expected to consume 2.2 gigawatts of power, surpassing Oracle's facility for OpenAI [3] - Anthropic, founded in 2021 by former OpenAI employees, has a recent valuation of $61.5 billion, with other investors including Microsoft and Salesforce Ventures [3] Group 3 - The partnership between Amazon and Anthropic is viewed as potentially more stable than the relationship between Microsoft and OpenAI, due to Anthropic's structure as a for-profit entity [5] - However, there are risks associated with Amazon's decision to invest in training its own AI models, which could impact Anthropic's reliance on Amazon for enterprise customer channels [5]

亚马逊正考虑向AI初创公司Anthropic追加数十亿美元投资，以深化两家公司在人工智能领域的战略联 盟。这项新投资将使亚马逊在已投入80亿美元基础上进一步加码，旨在帮助双方在全球AI商业化竞争 中获得优势。 7月10日，据媒体报道，知情人士透露，总部位于西雅图的云计算和电商巨头亚马逊已讨论扩大对AI模 型构建商Anthropic的投资计划。新的投资将确保亚马逊保持Anthropic最大股东之一的地位，与已投资 超过30亿美元的谷歌展开竞争，同时抗衡微软与OpenAI之间类似的数十亿美元合作关系。 这项投资还将深化双方在全球最大数据中心项目之一上的合作，并加强向亚马逊云计算客户销售 Anthropic技术的团队协作。亚马逊全球企业发展副总裁Dan Grossman表示："我们很快意识到双方有许 多根本性的共同目标。现有投资的规模代表了我们的雄心。" 不过，报道指出，两家公司的紧密结盟策略并非没有风险。该战略联盟面临的风险在于，微软对 OpenAI的140亿美元投资虽然帮助两家公司在AI产品商业化竞赛中取得早期领先，但由于ChatGPT制造 商希望转向营利模式，这一联盟正面临压力。 史无前例的数据中心合作 今年 ...

| Company | | Location | Number Of Chips | Type Of Chip | Status | Notes | | --- | --- | --- | --- | --- | --- | --- | | | | U.S. | 16,384 | H100 | Operating | Nvidia rents these servers for its DGX Cloud service | | awa | Amazon Web Services | Berwick, Pa. | Unknown | GPU | Planned | Data centers would be located | | | | | | | | next to a nuclear power plant | | | | Phoenix area | Unknown | GPU | Operating | | | | | U.S. | >200,000 | Trainium2 | Planned | AWS plans to build a Trainium | | | | | | | | clust ...

◼ AI ASIC 趋势已明朗，关注数据中心铜连接市场。 从目前已获知的 CSP ASIC 服务器方案来看，使用铜缆进行短距互 连已成趋势。根据 Fibermall 报告，AWS 今年有望采购 150 万颗自研芯 片，大部分使用AEC互连，尽管Trainium2的计算能力低于英伟达H100， 但可以使用 400G AEC，随着 Trainium3 于年底推出，800G AEC 的需求 有望得到增加；虽然微软在使用 AEC 铜缆方面较 AWS 缓慢，但目前已 经开始使用 AEC 构建 AI 网络；谷歌 TPU 互连目前主要使用无源铜缆 DAC，随着速度的提升，可能会转向 AEC。尽管目前还没有 Meta 使用 AEC 铜缆的明确信息，但在其 Minerva 机柜系统中，使用了 2 个可拆卸 电缆背板盒来连接计算托盘以及网络托盘。该电缆背板盒使用了 112G PAM4 连接器和电缆，每个盒子包括 4 个电缆组，每个电缆组包含 384 对不同电缆线，在 8 个 MTIA 托盘，6 个网络托盘和 1 个机柜管理系统 托盘间形成了 1536 对电缆网络。头部 CSP 厂商之外，X.AI 也对 AEC 表现出了大量需求， ...

2、模型阵营分明！aws不主推 openai gpt系列模型，不上Gemini系列模型室。google云上claude ，不上 gpt系列，微软云补主推 claude。 3、trainium2 目前能搭6万卡集群，后续推理需求，tainium2 推广很猛！inferentia好久没跌代了，后续估 计不会再推了。年底出trainium3。 最近抽空参加了一下现场。 说一下结论： 1、Claude 3.7 和4出来后，目前已经和openai O1 系列模型分庭抗礼了。 从单日token看，基本快持平 了。 4、亚马逊规模最大，基于CPU的计算 基础云，它受认可度最高。并在不断降本。 5、对应用开发递推有三层：基于gpu的 sagemaker！基础模型api调用一体化平台bedrock！针对 高阶用 户的Q！ ...

Hutt表示，虽然英伟达的Blackwell芯片在性能上优于Trainium2，但AWS芯片在性价比上更具优势。 "Trainium3将于今年推出，其性能是Trainium2的两倍，并将在节能方面再提升50%。"他说。 AWSAnnapurnaLabs工程总监RamiSinno表示，目前市场对这些芯片的需求已经超过供应。 "我们的供应能力非常强大，但我们开发的每一项服务背后都已经有客户在排队等着用了。"他说。 亚马逊云服务（AWS）即将宣布其Graviton 4芯片的一项升级：将网络带宽提升至每秒600Gbps，这也 是AWS称之为"公有云中最高规格"的配置。 AWS的杰出工程师AliSaidi将这一速度比作"一台机器每秒可读取100张音乐CD"。 Graviton 4是一款中央处理器（CPU），由亚马逊(AMZN.O)位于德州奥斯汀的Annapurna Labs设计开 发，是亚马逊定制芯片战略的一部分，用于与英特尔(INTC.O)和AMD(AMD.O)等传统半导体巨头竞 争。 但真正的战场在于人工智能基础设施领域，与英伟达的正面对决。 在AWS去年12月的re:Invent2024大会上，亚马逊宣布了Pro ...

智通财经APP获悉，亚马逊(AMZN.US)旗下云计算服务平台AWS将宣布对其Graviton4芯片进行更新， 该芯片的网络带宽达到每秒600千兆比特，该公司称这是公共云中的最高带宽。 AWS工程师Ali Saidi将这种速度比作一台机器每秒读取100张音乐CD。 Graviton4是一款中央处理器(CPU)，是亚马逊位于德克萨斯州奥斯汀的Annapurna实验室推出的众多芯 片产品之一。这款芯片是该公司定制战略的一大胜利，使其能够与英特尔(INTC.US)和AMD(AMD.US) 等传统半导体厂商展开竞争。 但真正的战斗是在人工智能基础设施领域挑战英伟达(NVDA.US)。 在去年12月举行的AWS re:Invent 2024大会上，该公司宣布了为初创公司Anthropic打造的人工智能超级 计算机Project Rainier。AWS已经投入80亿美元支持Anthropic。 根据AWS的说法，Anthropic的Claude Opus 4人工智能模型是在Trainium2 GPU上启动的，而Project Rainier则由超过50万块芯片驱动——这笔订单传统上会落入英伟达手中。 Hutt表示，虽然英 ...

2/25 11/24 沪 深 300 行业评级:增持 报告日期:2025-06-05 行业指数与沪深 300 走势比较 电子 行业点评 博通推出首款 102.4Tbps 超级芯片，AI 基础设施加速建设 分析师:陈耀波 执业证书号：S0010523060001 邮箱: chenyaobo@hazq.com 主要观点: ● 博通推出首款 102. 4Tbps 超级芯片,是目前市场上以太网交换机带 宽的两倍 博通宣布推出最新一代数据中心交换机芯片 Tomahawk 6. 这也是全球 首款单芯片交换容量达 102. 4Tbps 的芯片。这一芯片的性能较现有以太 网交换机翻倍,专为 AI 时代设计,可支持超大规模 GPU 集群的互联需 求, 单颗芯片即可驱动 10 万张 GPU 协同工作。 Tomahawk 6 的带宽理论峰值达 102Tbps. 相当于每秒处理 2.5 万部 4K 电影,较前代 Tomahawk 5 芯片实现６倍吞吐能力提升。是目前市场上 以太网交换机带宽的两倍。 从博通公告获悉,凭借前所未有的可扩展性、能效和 AI 优化功能, Tomahawk6 专为下一代可扩展和可扩展 AI 网络而设计,通过支 ...

Core Viewpoint - Amazon's stock has seen significant recovery, doubling from around $84 to $200, driven by improved financial performance and growth in various segments [1] Financial Performance - In 2024, Amazon's revenue increased by 11% to $638 billion, with North America up by 10%, International up by 9%, and AWS up by 19% [2] - Operating profit surged by 86% from $36.9 billion to $68.6 billion, indicating strong cost management and operational efficiency [3] Operational Improvements - Amazon enhanced delivery speeds, delivering over 65% more items to Prime members the same day or overnight compared to Q4 2023 [4] - The company launched Amazon Haul, a new ultra-low-price shopping service to compete with low-cost rivals [4] Cloud Computing and AI - AWS made significant advancements, including the introduction of the Trainium2 AI chip and new features in Amazon Bedrock, positioning the company favorably in the AI sector [5][6] - The global AI market is projected to grow from $294 billion in 2024 to $1.772 trillion by 2032, with AWS holding a 30% market share [8] Advertising Business Growth - Amazon's advertising segment generated $14 billion in revenue in Q1 2025, up 18% year-over-year, indicating strong growth potential [9] E-commerce Outlook - The e-commerce segment may experience mixed performance, with opportunities to gain market share but facing competition from emerging players [10] Company Culture - Amazon's "Day 1" mentality emphasizes customer focus, innovation, and agility, which is crucial for maintaining competitiveness [11] Stock Valuation - Amazon's current price-to-sales (P/S) ratio is 3.3, which is within its historical range, suggesting the stock is neither a bargain nor excessively priced [12] Investment Implications - Strong results in 2024 highlight Amazon's execution capabilities, with AWS and advertising expected to drive future growth [13] - While the stock is not a strong buy, it is advisable for existing investors to hold, and long-term investors may consider gradually building a position [14]