OpenAI转向谷歌TPU芯片，AI基础设施格局的重要转折点。 据追风交易台的消息，摩根士丹利在6月30日的报告中表示，对谷歌而言，这是OpenAI对其AI基础设施 能力的重要背书，将推动谷歌云业务增长并巩固其在ASIC生态系统中的领先地位。 对英伟达而言，虽然面临产能约束挑战，但仍将保持GPU市场主导地位。对亚马逊而言，OpenAI合作 伙伴名单中缺少AWS可能暴露其产能限制和Trainium芯片竞争力不足的问题。 OpenAI首次大规模采用非英伟达芯片 华尔街见闻之前消息，OpenAI近期开始租用谷歌的TPU芯片为ChatGPT等产品提供算力支持，这是该公 司首次大规模使用非英伟达芯片。此次合作安排使得OpenAI缓解对微软数据中心的依赖，同时为谷歌 的TPU提供了挑战英伟达GPU市场主导地位的机会。OpenAI希望通过谷歌云租用的TPU芯片降低推理 计算成本。 摩根士丹利认为，OpenAI是迄今为止最重要的TPU客户（其他客户包括苹果、Safe Superintelligence和 Cohere），这一协议是对谷歌AI基础设施能力的重要认可。谷歌的TPU技术已经发展了十年，最初的 TPU于2015年发布 ...

摘要 光模块市场景气度持续上修，受益于互联网云厂商规划新发展及乐观的 投关交流，预计 2026 年 800G 和 1.6T 光模块需求量将分别增至 4,000 万只和 600-800 万只，主要驱动力来自 CSB 厂商架构变化和市 场需求增加。 旭创保持市场领先地位，新易盛份额快速提升，与 Meta、亚马逊等合 作紧密，并积极拓展英伟达等客户。预计 2025 年龙头公司收入可达 360-370 亿人民币，2026 年达 440-450 亿人民币，净利润率提升潜 力巨大。 二线光模块厂商如索尔思、联特科技和剑桥科技有望分享市场份额，行 业贝塔不断上修。国内互联网音响加大投入，华工科技和光迅科技等国 内龙头企业受益，预计 2025 年下半年业绩强劲。 光模块跳线市场潜力巨大，800G 光模块跳线需求量约为 2000 万支， 市场空间达数百亿人民币。泰胜光博创、亨通光电、爱康得等厂商有望 受益，业绩弹性大。 Q&A 近期光模块行业的市场表现如何？ 近期光模块行业表现突出，尤其是在 6 月份通信行业中排名第一。头部公司如 旭创、新易盛、太辰光和博创的涨幅显著，达到百分之四五十。整个互联网云 场对 ASEC 的重视程度 ...

Core Viewpoint - The article discusses the increasing competition in the AI chip market, particularly how major tech companies like Google and Meta are accelerating their development of custom chips to reduce reliance on Nvidia's GPUs, with predictions that ASIC shipments will surpass Nvidia's AI GPU shipments as early as next year [1][2]. Group 1: Market Dynamics - Nvidia has historically dominated the AI chip market, holding over 80% market share in AI servers, while ASIC-based servers currently account for only 8% to 11% [2][4]. - Google is expected to ship between 1.5 million to 2 million of its self-developed AI chips (TPUs) this year, while Amazon's AWS is projected to ship 1.4 million to 1.5 million ASICs, bringing their combined shipments close to half of Nvidia's estimated annual GPU shipments of 5 million to 6 million [2][4]. Group 2: Cost Efficiency - The key advantage driving tech giants to develop their own chips is the reduction in Total Cost of Ownership (TCO), with ASICs potentially saving 30% to 50% in TCO compared to GPUs [3]. - Google claims its TPUs can deliver three times the performance of Nvidia GPUs per unit of energy consumed, highlighting the efficiency of custom chips [3]. Group 3: Competitive Landscape - Meta is focusing on launching its new high-performance ASIC chip "MTIA T-V1" in Q4 of this year, aiming to outperform Nvidia's next-generation AI GPU "Rubin" [5]. - Despite ambitious plans, Meta faces production challenges due to limited advanced packaging capacity from TSMC, which can only provide 300,000 to 400,000 units, creating a bottleneck [5]. Group 4: Nvidia's Response - In response to the competitive threat, Nvidia has opened its proprietary "NVIDIA NVLink" communication protocol to facilitate integration with other companies' CPUs or ASICs, aiming to retain its major clients [6]. - Nvidia's established software ecosystem, CUDA, remains a significant barrier for competitors, as it allows AI developers to efficiently build and deploy applications, maintaining Nvidia's competitive edge [6].

Core Viewpoint - 嘉楠科技 is terminating its non-core AI semiconductor business to focus on its core operations, which include Bitcoin mining machine sales, self-mining, and consumer mining products. This strategic shift is expected to significantly reduce operational costs in the coming months [1][2]. Group 1: Business Strategy - The decision to end the AI semiconductor business is primarily due to its low revenue of approximately $900,000 for FY 2024, which accounts for 15% of the company's total operational expenses, thus dragging down overall efficiency [2]. - Post-strategic adjustment, resources will be concentrated on mining equipment sales, self-mining, and consumer-grade mining machines, where the company has established a clear competitive advantage [2]. Group 2: Mining Operations - 嘉楠科技's mining business is set to launch the Avalon Q mining machine in 2025, which will be compatible with global voltage standards, and the next-generation A16 series is expected to achieve an energy efficiency ratio of 15J/TH, nearing industry-leading levels [2]. - The company has a global hash rate of 8.15 EH/s in North America, with a low average electricity cost of $0.044 per kWh, and overseas projects in Ethiopia are operating with over 95% uptime [2]. Group 3: Financial Performance - In Q4 FY 2024, 嘉楠科技 reported a 33.2% reduction in net losses and achieved positive EBITDA, indicating improved financial health [2]. - The Avalon miner series, particularly the A14 product line, saw significant revenue growth following the Bitcoin halving event in April 2024, with demand for the A15 and upcoming A16 series continuing to rise [3]. Group 4: Company Milestones - 嘉楠科技 was founded in 2013 and released the world's first blockchain computing device based on ASIC chips, marking the beginning of the ASIC era in the industry [4]. - The company underwent a name change in 2015 and achieved significant milestones in chip production, including the mass production of 28nm and 16nm chips [5][6].

May 2025 NASDAQ: BTDR Disclaimer & Forward-Looking Statements This presentation and the accompanying oral presentation have been prepared by Bitdeer Technologies Group (the "Company") solely for information purpose, and has not been independently verified. By viewing or accessing the information contained in this presentation, the recipient hereby acknowledges and agrees that no representations, warranties or undertakings, express or implied, are made by the Company or any of its directors, shareholders, em ...

本文来自微信公众号：经济观察报观察家 （ID：eeoobserver），作者：陈永伟，原文标题：《能否温 柔地跨过那个"奇点"》，题图来自：AI生成 6月10日，OpenAI首席执行官山姆·奥特曼（Sam Altman）在博客上发表《温柔的奇点》（The Gentle Singularity），并称这可能是他"最后一篇不依赖AI写成的文章"。他在文中表示，人类或许已跨越了通 往数字超级智能的"事件视界"，即进入了一个不可逆的AI发展阶段。 "奇点"原是数学术语，后被美国作家弗诺·文奇引入科技史领域，用以描述技术增长加速到无法预测的 临界点。未来学家雷·库兹韦尔则将其与AI相连，提出2045年将实现AI超越人类智能的"技术奇点"。近 年来，随着AI突飞猛进，这一预测不断被提前——库兹韦尔在其新书中将时间表改为2029年。 尽管"奇点"何时到来长期存在争议，许多人仍将其视为遥远的未来。然而奥特曼却宣称，它已经悄然降 临，并非以剧烈冲击，而是以温柔、几近无声的方式实现。这番言论引发热议：有人称其洞见深刻，有 人批评其过于乐观，亦有声音指出，这番言论发布时机恰逢GPT-o3pro亮相，或许不过是营销策略。 那么，奥特 ...

◼ 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， ...

1) AWS Trainium: it remains a global debate whether Marvell will win part of the Trainium3 business (link). At least our checks suggest that Alchip taped out the Trainium3 design back in February, and the wafers were already out in May. Charlie Chan thinks Alchip has a higher chance to win the 2nm Trainium4, which should be decided this summer. Astera Labs and Alchip just formed a partnership in connectivity chip design (I/O chiplets). We think that may help Alchip compete for next-generation XPU ASIC proje ...

Asia AI Server Global Markets Research EQUITY: TECHNOLOGY Meta's ambition on ASIC AI server Meta's MTIA AI servers could mark a milestone in 2026F AI ASIC is booming; a potential cross over in units from 2H26-2027 In the world of AI servers, nVidia (NVDA US, Not rated) has been dominating 80%+ market "value" so far, vs. ASIC AI servers that have roughly 8-11% in value share (based on Bloomberg consensus estimates). However, if we just compare the number of units for AI ASICs vs. nVidia's AI GPUs (which can ...

Core Viewpoint - Marvell Technology (MRVL.US) has seen a significant stock price increase due to positive evaluations from top Wall Street analysts regarding its customized AI ASIC chip activities and potential market announcements [1][3] Group 1: Market Opportunities - Analysts from Evercore ISI predict that the new AI ASIC chip designs could ramp up quickly between 2026 and 2027, indicating strong future demand [1] - Marvell expects each customized AI chip design win to generate billions in lifecycle revenue within 1.5 to 2 years, while each XPU Attach win could contribute hundreds of millions within 2 to 4 years [2] - The total addressable market (TAM) for customized data center chips has been raised to $94 billion, a 26% increase from last year's AI activities [3] Group 2: Financial Projections - Marvell has raised its financial targets, with analysts noting that the potential earnings per share could reach $8 by 2028, exceeding Wall Street estimates by 60% [4] - The company aims to capture at least 20% of the TAM, with over 50% of its data center revenue expected to come from AI ASIC-related demands [3][5] Group 3: Competitive Landscape - Broadcom (AVGO.US) is identified as the long-term beneficiary of Marvell's AI activities, holding a dominant market share of approximately 60% in the AI ASIC sector, while Marvell holds 13% to 15% [6][7] - The AI ASIC market is expected to grow significantly, with major tech companies like Google, Microsoft, and Amazon investing heavily in AI ASIC chips, indicating a shift in market dynamics away from GPU dominance [7]