行业研究 · 行业专题 通信 投资评级：优于大市（维持评级） 2026年02月26日 证券研究报告 | 2026年02月26日 数据中心互联技术专题： AI变革推动OCS新技术快速发展 国信通信·行业专题报告 证券分析师：熊莉 联系人：赵屿 021-61761067 021-61761068 xiongli1@guosen.com.cn zhaoyu6@guosen.com.cn S0980519030002 请务必阅读正文之后的免责声明及其项下所有内容 投资摘要 光交换机 OCS (Optical Circuit Switch) 是一种无需光电/电光（O/E/O）转换，直接实现光信号在光纤端口间切换的技术，应用于 AI 算力集群、超大规模数据中心的叶脊架构互连、超节点集群高速通信等场景。通过直接在光域完成数据信号的路由与切换，无需进行 光电 / 电光转换，OCS 技术从底层规避了传统电交换在高速传输下的带宽瓶颈、功耗损耗问题，大幅缩短了信号传输的时延，同时其功 耗仅与端口数相关、与信号传输速率无关，大大降低减少了功耗。行业技术研究与落地实践验证，OCS 技术可助力 AI 算力集群、数据 中心光互连系统整体功 ...

6. Alphabet将机器人软件子公司Intrinsic重新划归至谷歌旗下，加码物理AI布局。 美股时段值得关注的事件： 1. 美国三大股指期货小幅上涨，纳指期货涨0.04%，标普500指数期货涨0.1%，道指期货涨0.11%。 2. 欧股主要指数齐涨，德国DAX指数涨0.51%，英国富时100指数涨0.12%，法国CAC指数涨0.95%，欧 洲斯托克50指数涨0.42%。 3. 英伟达Q4营收同比增长73%至681亿美元创新高，Q1业绩指引强劲；预期芯片收入将超越5000亿美元 的目标。 4. 英伟达CFO：公司获准向中国出口少量H200芯片，但尚未获得任何收入。 5. 据报苹果为iPhone 17系列敲定三星DS部门LPDDR5X订单，价格提高100%。 21:30 美国至2月21日当周初请失业金人数 (格隆汇) 7. 特斯拉中国"变相降价"，推出五年零息购车方案。 8. 礼来：在2型糖尿病试验中，Orforglipron在血糖控制和减重方面均优于司美格鲁肽。 9. AMD和Nutanix联合开发一个开放的全栈人工智能(AI)基础设施平台。 10. SK海力士携手闪迪启动高频宽快闪记忆体全球标准化进程。 ...

The race to build AI infrastructure has led to a significant financial overhang. These tech giants have entered into long-term data center lease commitments that have yet to commence and, therefore, are not classified as current liabilities. As a result, the obligations do not appear on their balance sheets under generally accepted accounting principles (GAAP), Fortune reported, citing a Moody's report released Monday.By 2025 end, the companies had accumulated $969 billion in total undiscounted future lease ...

Tesla Inc.‘s (NASDAQ:TSLA) Robotaxi in Austin is cheaper when compared to its rival,  Alphabet Inc.'s (NASDAQ:GOOGL) (NASDAQ:GOOG) Waymo.2x Cheaper Than RivalIn a new research note released on Thursday, analysts from investment bank Jeffries analyzed ride-hailing services from Tesla and Waymo, taking over 15 rides in a Tesla and over 19 rides in Waymo. Analysts found that Tesla offered just 2 of the 15 rides without a safety driver and that they also failed to book a ride due to unavailability over 25% of t ...

2026 年 02 月 26 日 ledi/229 相关研究 证券分析师 林起贤 A0230519060002 linqx@swsresearch.com 袁伟嘉 A0230519080013 yuanwj@swsresearch.com 任梦妮 A0230521100005 renmn@swsresearch.com 夏嘉励 A0230522090001 xiajl@swsresearch.com 赵航 A0230522100002 zhaohang@swsresearch.com 黄俊儒 A0230525070008 huangjr@swsresearch.com 李国盛 A0230521080003 ligs@swsresearch.com 研究反馈 张淇元 A0230124080001 zhangqy@swsresearch.com 联系人 张淇元 A0230124080001 zhangqy@swsresearch.com Seedance2.0 突破, Al 视频 格局及产业链机遇 ——GenAl 系列报告之 69 暨 Al 应用深度之四 本期投资提示: 请务必仔细阅读正文之后的各项信息披露与声明 ...

Core Insights - The AI glasses market in 2026 is characterized by intense competition, with numerous companies, including major tech giants and startups, launching new products and strategies, marking a shift from niche gadgets to essential tools for various applications [1][2][4] Group 1: Market Dynamics - The AI glasses market is witnessing a "full-scale war" with major players like Meta, Google, Apple, and domestic companies like Huawei and ByteDance all actively participating [1][2] - The CES 2026 event served as a platform for companies to showcase their innovations, with significant product announcements from various manufacturers [2][4] - The trend of "lightweight" and "independence" is evident, as companies prioritize user comfort and reduce reliance on smartphones through features like eSIM and independent processing capabilities [9][10] Group 2: Product Launches and Innovations - Key product launches include: - XGIMI's MemoMind brand entering the smart glasses market [8] - Thunderbird's X3 Pro Project eSIM, the first dual-lens AR glasses with independent connectivity [8] - Rokid's ultra-light AI glasses weighing only 38.5 grams, emphasizing comfort and audio interaction [8] - Modular AR glasses M6 from Xuanjing, designed for hardware upgrades [3][8] - Meta plans to enhance its Ray-Ban Meta glasses, aiming for improved optical comfort and AI interaction [4][8] - Google's three-step strategy includes launching audio smart glasses and collaborating on Project Aura with XREAL [4][8] - Apple is expected to release Apple Glasses by the end of 2026, focusing on a lightweight design and AI integration [4][8] Group 3: Industry Trends and Drivers - The industry is driven by three main factors: cost reduction in the supply chain, advancements in AI interaction, and growing market demand [10][11] - The rapid decrease in component costs, particularly in chips and optical displays, has made AI glasses more affordable for consumers, with prices now in the 2000-3000 yuan range [11] - The introduction of national subsidies for smart glasses is expected to boost market penetration and consumer adoption [14][15] Group 4: Future Competition Landscape - The competition is shifting from hardware specifications to user experience, ecosystem integration, and technological innovation [16][17] - Companies are focusing on creating compelling reasons for daily use, with examples like Rokid's integration of AI glasses into automotive systems and educational tools [17][18] - The future of the industry will depend on ecosystem development, with companies like Google and Huawei aiming to create comprehensive platforms that integrate AI capabilities and user data [20][21]

Core Insights - The investment wave in AI infrastructure is pushing tech giants into an unprecedented capital-intensive cycle, with hyperscalers like Amazon, Google, Meta, Microsoft, and Oracle expected to exceed historical capital expenditure peaks from the internet bubble era [1][4]. Capital Expenditure Trends - Morgan Stanley forecasts that the capital expenditure to sales ratio for these hyperscalers will reach 34%, 39%, and 37% from 2026 to 2028, surpassing the internet bubble peak of approximately 32% [1][4]. - Including financing leases, this ratio could rise to 38%, 44%, and 45% during the same period [1][7]. - The total capital expenditure for these companies is projected to exceed $2 trillion over the next three years, accounting for about 40% of the total capital expenditure of Russell 1000 index constituents [1]. Revenue Adjustments - Despite the significant increase in capital expenditure, revenue forecasts have not seen a corresponding rise, leading to a decline in free cash flow (FCF) expectations for hyperscalers [3][10]. - Over the past six months, capital expenditure expectations for 2026 and 2027 have been raised by over $630 billion, while revenue expectations have only seen limited adjustments [3][10]. Financing Leases Impact - The use of financing leases has significantly inflated the actual investment scale, with total commitments for future leases exceeding $660 billion among the five companies [13]. - For instance, Oracle's capital expenditure to sales ratio could rise dramatically from 75% to over 100% when including financing leases [15]. Sector Performance Disparity - Semiconductor companies have emerged as the biggest beneficiaries of the current investment cycle, with their revenue expectations rising by approximately 60%, compared to only 8% for hyperscalers [3][17]. - The market has shown a preference for semiconductor firms, with stock prices increasing significantly more than those of hyperscalers since December 2023 [17]. Future Outlook - Analysts believe that while companies like Meta, Google, and Amazon are leveraging AI investments to enhance user engagement and monetization, the substantial capital expenditures will lead to increased depreciation costs, putting pressure on profit margins if sales do not keep pace [18].

Core Insights - The "Magnificent Seven" companies, including Nvidia, Apple, Alphabet, Microsoft, Amazon, Meta Platforms, and Tesla, have been pivotal in driving the stock market higher, particularly due to advancements in artificial intelligence [1][4] - However, these companies are collectively signaling a concerning trend with over $8.4 billion in net stock sales by insiders over the past year [4][10] Insider Trading Activity - Insiders at the Magnificent Seven have been significant net sellers of their company's shares, with the exception of Tesla, where CEO Elon Musk purchased approximately $1 billion in shares [5][10] - The total net selling across these companies amounts to $8,414,225,611, with Nvidia leading at $2,249,988,017 and Amazon following with $5,724,146,695 [10] - Notably, there has been a lack of insider buying in major companies like Nvidia, Apple, Amazon, and Meta Platforms, raising potential concerns for investors [12][13] Market Context - The insider trading data indicates that while selling may not always be negative, the absence of insider buying amidst a historically expensive stock market could be a red flag [9][12] - Insiders at Microsoft and Alphabet have made minimal purchases, totaling $3.4 million and $5 million respectively, which is relatively low compared to the selling activity [13]

Group 1 - Samsung's latest Galaxy S26 series enhances integration with Gemini, introducing the Gemini smartphone assistant to automate tasks for users, similar to domestic "Doubao phone assistant" [1] - Users can activate Gemini by long-pressing the side button, allowing it to serve as a "to-do task" entry point, executing multi-step operations while users continue to use their phones and monitor progress via the notification bar [1] - Some task execution capabilities of the Gemini assistant are realized through GUI solutions [1] Group 2 - Google’s Android ecosystem president, Sameer Samat, stated that Android is evolving from a traditional operating system to a "truly understanding and serving intelligent system" [3] - He referred to the Gemini smartphone assistant as "agentic AI," emphasizing its ability to "get things done" by reducing app switching and manual coordination, thus streamlining processes while maintaining user visibility and control [3] - Samat highlighted the emerging nature of this technology and the importance of embracing it [3]

IMO金牌已经"过时"了。 基于Gemini 3 Deep Think的谷歌数学智能体Aletheia在更难的挑战赛FirstProof中拿下的最佳成绩。 在公布的完整成绩单中，10道题Aletheia全程0人工参与解出6道，其中5题专家全票通过，还有一题拿到了5/7的通过率。 | | Aletheia | Expert Evaluation | | --- | --- | --- | | | (best of 2) | (correct/total) | | P1 | N/A | | | P2 | Correct | 4/4 | | РЗ | N/A | | | P4 | N/A | | | P5 | Correct | 4/4 | | be | N/A | | | P7 | Correct | 3/3 | | P8 | Correct? | 5/7 | | P9 | Correct | 4/4 | | P10 | | 2/2 | FirstProof是由来自哈佛、斯坦福等名校的11位顶尖数学家联手打造的一套专门验证AI独立科研能力的数学题集。 10道题全网无迹可循，没法儿背答案作弊，连陶哲轩都转发说这事儿 ...