
2026年9月9日 星期三
NO.1 亚马逊重金押注高通AI芯片:最高600亿美元采购额与2500万股认股权证深度绑定
9月8日,亚马逊与高通正式宣布达成一项多代际合作,双方携手为AI数据中心研发定制化的AI推理芯片,并推出最高支持1.6T的光互联解决方案。根据双方协议,高通将向亚马逊发行至多2500万股普通股认股权证,设定的行权价格为每股161.26美元。这些认股权证的归属条件与亚马逊实际采购高通服务器芯片的金额直接挂钩,整体采购规模上限高达600亿美元,且已有375万股因亚马逊的初始采购承诺而完成归属。
点评:这项合作标志着高通正加速将业务版图从移动端向AI数据中心延伸。与此同时,亚马逊也通过定制芯片与高速互联技术进一步夯实了AWS自有的AI基础设施,双方在供应链层面实现了更深层次的相互绑定。
NO.2 OpenAI企业业务7月年化营收环比大增32%,首席财务官直言“势头强劲”
当地时间9月8日,OpenAI首席财务官Sarah Friar在高盛举办的会议上透露,公司企业业务近期呈现出“势头强劲”的发展态势。数据显示,7月份企业业务的年化营收运行率环比攀升32%,而同期公司整体年化营收运行率的增幅为20%。她指出,尽管消费者业务同样保持着稳健的增长步伐,但企业业务的增长尤为亮眼。今年4月,OpenAI曾透露企业业务收入在总营收中的占比已突破40%,并预计至2026年底,该板块规模将与消费者业务基本持平。
点评:企业业务增速大幅跑赢整体业务,这表明OpenAI的商业化进程正逐步从单纯的消费者订阅,向企业订阅、API调用以及Agent工作流等方向深度拓展。未来,企业客户极有可能成为驱动公司营收持续增长的核心引擎。
NO.3 Meta发布个人AI智能体Muse,扎克伯格视其为最大商业契机
当地时间9月8日,Meta面向美国用户推出了个人AI智能体Muse。该智能体能够代为处理网购、买票、预约球课、填写表单等各类日常事务,并已打通Gmail、Google日历、OpenTable、Instagram及WhatsApp等多项服务。Meta首席AI官Alexandr Wang强调,Muse的根本宗旨是让普通用户也能轻松上手。此前,扎克伯格曾公开表示,个人AI助手是Meta面临的最大商业机遇,公司的终极目标是将其触达到Instagram、Facebook和WhatsApp庞大的数十亿用户群。目前,Muse提供了免费版以及定价为20美元和100美元每月的付费订阅选项,后续还计划与Meta AI眼镜进行整合。
点评:Meta正致力于推动AI从单纯的聊天工具,进化为能够直接代办任务的个人智能体。其庞大的数十亿社交用户基数及成熟的应用生态,有望成为其在商业化落地过程中的核心优势。
NO.4 谷歌TPU每美元性能最高超英伟达B200达50%,外部化步伐提速
半导体研究机构SemiAnalysis于9月7日发布的最新测试结果显示,在部分特定推理场景下,谷歌TPUv7 Ironwood的每美元性能表现最高比英伟达B200高出50.4%,其每百万Token的推理成本最低仅需约0.098美元。伴随Anthropic等外部客户逐步增加采购量,谷歌正加快TPU外部化的进程。此外,谷歌还计划在10月份开源TorchTPU,通过原生PyTorch后端来降低开发人员使用TPU的难度;而下一代TPUv8i更是将支持原生FP4计算。
点评:谷歌正通过芯片硬件、网络互联与软件生态的协同效应来不断降低TPU的应用门槛。未来,若外部客户规模持续扩大且软件生态愈发完善,TPU有望突破谷歌内部算力的局限,成长为英伟达AI推理芯片领域的重要竞品。
NO.5 康宁携手Verizon签订价值数十亿美元的光纤长期大单
当地时间9月8日,康宁与美国电信巨头Verizon正式签订了一份延续至2032年的多年期光纤供应大单,总价值达数十亿美元。根据规划,在2027年至2032年期间,康宁将提供累计超过8000万英里的高密度光纤及配套连接方案,以服务于家庭宽带网络及AI数据中心骨干网的建设。此前,康宁已相继与Meta、英伟达、亚马逊等客户签署了多项大额长期供货协议。该公司指出,AI数据中心单个节点所需的光纤用量大约是传统交换机的16倍之多。
点评:在AI数据中心建设浪潮的带动下,骨干网络正迎来扩容周期,光纤需求正从云端巨头和芯片厂商向电信运营商蔓延。这种多年期的长期订单也清晰地折射出产业链上下游提前锁定产能的普遍趋势。
免责声明:本文所提供的内容及数据仅供交流参考,并不构成任何投资建议,请读者在使用前自行核实。若据此进行操作,相关风险需自行承担。
本文来源:https://www.sohu.com/a/1073615484_115362
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
Wednesday, September 9, 2026
NO.1 Amazon Makes Massive Bet on Qualcomm AI Chips: Procurement Cap of $60 Billion Linked to 25.5 Million Shares
On September 8, Amazon and Qualcomm announced a long-term partnership to develop custom AI inference chips for data centers, alongside a 1.6T optical interconnect solution. Qualcomm will issue up to 25.5 million shares to Amazon, with an exercise price of $161.26 per share. The vesting of these shares is directly tied to Amazon's actual procurement of Qualcomm server chips, with a total procurement limit of $60 billion. Already, 3.75 million shares have been vested due to Amazon's initial purchase commitment.
This collaboration marks a significant expansion of Qualcomm's business from mobile to AI data centers. For Amazon, it solidifies its AWS AI infrastructure with bespoke chips and high-speed interconnects, creating a deep supply chain interdependence between the two companies.
NO.2 OpenAI's Enterprise Business Annual Revenue Soars 32% Month-over-Month in July, CFO Declares "Strong Momentum"
On September 8, OpenAI CFO Sarah Friar revealed that the company's enterprise business is experiencing "strong momentum," with a 32% increase in annualized revenue run rate from June to July. In contrast, the total company's annualized revenue grew by 20% over the same period. While consumer business growth remains stable, the enterprise sector is surging, already accounting for over 40% of total revenue and expected to match consumer revenues by 2026.
The significant outperformance of the enterprise sector indicates a shift in OpenAI's revenue drivers from simple consumer subscriptions to larger-scale enterprise deals and API usage. The enterprise market is poised to become the primary engine for the company's revenue growth.
NO.3 Meta Launches Personal AI Agent Muse, Viewed by Zuckerberg as the Biggest Commercial Opportunity
On September 8, Meta unveiled Muse, a personal AI agent for US users capable of performing daily tasks such as online shopping, ticketing, booking sports classes, and filling out forms. Muse integrates with services like Gmail, Google Calendar, OpenTable, Instagram, and WhatsApp. Meta's chief AI officer, Alexandr Wang, emphasized that Muse is designed to be accessible to everyone. CEO Mark Zuckerberg has previously stated that personal AI assistants represent the company's largest commercial opportunity, aiming to reach billions of users across Meta's platforms. Muse is available in free and paid subscription tiers ($20 and $100 per month) and will eventually integrate with Meta's AI glasses.
Meta is evolving its AI offerings from simple chat bots to task-performing agents. Given its massive user base and established application ecosystem, Meta is well-positioned to capitalize on this transition.
NO.4 Google TPU Performance per Dollar Surpasses NVIDIA B200 by 50%, Accelerating Externalization
A report by SemiAnalysis published on September 7 shows that Google's TPUv7 Ironwood achieves a cost-performance ratio up to 50.4% higher than NVIDIA's B200 in certain inference scenarios, with costs as low as $0.098 per million tokens. As external customers like Anthropic increase their purchases, Google is accelerating the externalization of its TPU technology. The company also plans to open-source TorchTPU in October and upgrade upcoming models to support native FP4 computation.
Google is leveraging hardware, networking, and software synergies to reduce the barrier to entry for TPUs. With a growing external customer base and an improving software ecosystem, TPUs have the potential to become a direct competitor to NVIDIA's AI inference chips.
NO.5 Corning Secures Multi-Billion Dollar Fiber Optic Contract with Verizon
On September 8, Corning signed a long-term agreement with Verizon to supply fiber optics until 2032, valued in the billions of dollars. The contract includes the delivery of over 80 million miles of high-density fiber and connectivity solutions from 2027 to 2032, intended for broadband networks and AI data center backbones. Corning has previously secured large contracts with Meta, NVIDIA, and Amazon. The company notes that AI data centers require significantly more fiber per node than traditional setups.
The demand for fiber optics is surging due to the expansion of networks for AI data centers. This long-term contract indicates that the demand is shifting from internet giants to telecommunications carriers, and the length of the agreement suggests a strategy to lock in future capacity.
Disclaimer: The contents and data in this article are for information purposes only and do not constitute investment advice. Readers should verify the information before making any decisions. Any actions taken based on this information are the reader's responsibility.
Source: https://www.sohu.com/a/1073615484_115362 If you have any copyright issues, please contact: 13951431913#139.com or leave a comment on the right.