证券研究报告 AI系列专题报告（三） AIoT端侧：智能硬件百花齐放，国产SoC大有可为 半导体行业强于大市（维持） 徐碧云 投资咨询资格编号：S1060523070002 闫磊 投资咨询资格编号：S1060517070006 徐勇 投资咨询资格编号：S1060519090004 证券分析师 2025年6月19日 请务必阅读正文后免责条款 投资要点 处理：边缘智能推动了NPU的广泛应用。AI端侧应用正加速渗透，技术革新推动硬件升级。音频作为高频次、高强度信息交互的重要载体，正快 速成为AI落地端侧的首要信息维度。AI增强型SoC通过集成NPU与可重构计算单元，结合算法-芯片协同优化，突破传统处理器的能效瓶颈，可有 效释放边缘侧的实时推理与决策能力。NPU凭借低功耗特性成为边缘设备的理想选择，专为神经网络设计的架构使其在深度学习任务中表现优异， 主要应用于人脸识别、语音识别、自动驾驶、智能相机等领域，且技术架构随AI算法和应用场景不断演进。 连接：无线通信，物联网主要实现方式。物联网端侧连接需求持续增长，推动局域无线连接技术的应用扩展，涵盖WiFi、蓝牙、ZigBee、2.4G 私有协议、Thread等 。 各 ...

机器之心报道 机器之心编辑部 人人都绕不开的推荐系统，如今正被注入新的 AI 动能。 随着 AI 领域掀起一场由大型语言模型（LLM）引领的生成式革命，它们凭借着强大的端到端学习能力、海量数据理解能力以及前所未有的内容生成潜力，开始重 塑各领域的传统技术栈。 作为互联网流量的核心引擎，推荐系统面临着级联架构导致的算力碎片化、优化目标割裂等问题，并逐渐制约其创新发展。实现从碎片化拼装到一体化整合的范 式跃迁，是推荐系统重焕生机的必由之路，而利用 LLM 技术重构架构以实现效果提升、成本降低成为关键。 近日，快手技术团队交出了他们的答卷，最新提出的「OneRec」首次以端到端生成式架构重构推荐系统全链路。 在效果与成本这场看似零和的博弈中，OneRec 让「既要又要」成为可能 ： 目前，该系统已在快手 App / 快手极速版双端服务所有用户，承接约 25% 的QPS（ 每秒请求数量 ） ，带动 App 停留时长提升 0.54%/1.24%，关键指标 7 日用户生 命周期（LT7）显著增长，为推荐系统从传统 Pipeline 迈向端到端生成式架构提供了首个工业级可行方案。 下图（左）展示了快手 / 快手极速版中 O ...

"强大记忆力"助力高效管理上千台服务器 随着企业云资源规模持续扩大，终端管理面临着多机器部署的难题。以服务器为例，据合合信息技术团队 人员介绍，在工作中，一名程序员一天需要同时应对四五十台服务器，遇到体量较大的项目，有的程序员 甚至要同时管理上千台服务器。当产品某个功能或者服务上线时，开发人员需要逐个将代码部署到成百上 千台服务器中，耗时长，效率十分低下。 面对大规模的服务器管理需求，与其他智能CLI Agent相比，Chaterm具备了批量管理远程服务器的独特能 力。通过自动"记忆"用户的操作习惯，即使没有ROOT权限，用户也可以在任意远程主机上实现个性化的语 法高亮或自定义的快捷命令，实现"一次配置，多端通用"的便捷体验。除此之外，Chaterm还具备了跨平台 兼容性，一键安装，支持MAC，WINDOWS，LINUX等操作系统，降低了企业混合IT环境下的运维管理复 杂度。 在数字化转型加速的当下，AI技术正深度重塑云计算产业。当企业将服务器、数据库等核心资产交由AI管 理时，数据安全问题成为重要考量。为了保护用户隐私，合合信息宣布全面开源Chaterm核心代码，开发者 可以直接观察算法底层运行逻辑，并根据 ...

图片来源： a16z Z Highlights Olivia Moore 是 Andreessen Horowitz 消费投资团队的合伙人，专注于 AI 领域； Marc Andrusko 是其 B2B AI 与金融科技方向的合伙人 。 两位合著者基于近 18 个 月对众多 AI 初创公司的追踪观察，在 2025 年 6 月 6 日分享了关于 AI 时代初创公司增长标准的实证发现。 增长的神话与现实 在生成式AI时代最常见的说法之一是："创业公司比以往任何时候都增长得更快"，而且常常是在资源更少的情况下实现的。根据公司公布的指标， Lovable 在短短六个月内就实现了5000万美元的收入 ， Cursor在第一年便报告了1亿美元的收入 ， Gamma在融资不足2500万美元的情况下实现了5000万美元的收 入 。 但对大多数普通的AI公司而言（而非顶尖0.1%），增长究竟意味着什么？在AI出现之前， 顶级企业级初创公司的常见衡量标准是，前12个月实现100万美 元ARR（年度经常性收入） 。相比之下，消费类公司通常会在第一年之后才开始盈利，通常要等到积累了数百万（甚至数千万）的用户基础后，才能通过 广告盈利 ...

Group 1 - A-shares opened slightly lower on June 19, with PCB and humanoid robot concept stocks leading the gains [1] - Semiconductor ETF Southern (159325) rose by 0.98%, while the Sci-Tech Chip ETF Southern (588890) increased by 0.73% [1] - Notable individual stocks included Juchip Technology, which surged over 12%, and Yangjie Technology, which rose over 5% [1] Group 2 - The China Securities Regulatory Commission (CSRC) issued opinions on enhancing the inclusiveness and adaptability of the Sci-Tech Innovation Board [1] - The CSRC aims to improve the coordination of investment and financing market functions and promote the inclusion of more long-term funds into new productive forces [1] - China International Capital Corporation (CICC) noted a structural increase in opportunities in the Sci-Tech sector, particularly in AI, robotics, and semiconductors [1] Group 3 - Huatai Securities highlighted three investment opportunities in the semiconductor equipment market for the second half of the year [2] - These opportunities include increased demand for advanced process logic and packaging driven by generative AI, structural opportunities from advanced process expansion in mainland China, and market share shifts due to U.S. export restrictions [2] - Investors without stock accounts can access the chip sector through the Sci-Tech Chip ETF Southern (588890) and related funds [2]

2025 年的 AI 浪潮席卷全球，从 DeepSeek 大模型的破圈效应到政策红利的持续加码，教育领域正在经历一场深刻的变革。当「个性化-高质量-大规模」 的教育困境被提出时，AI 的出现似乎成为破局的关键钥匙。 头部的教育科技公司纷纷接入 DeepSeek，结合 AI 研发产品、升级技术。2025 年，AI+教育的市场规模预计突破 700 亿元，已然有形成商业、教育的双重 闭环之势。 那么，这场变革风起何处？在新时代教育如何拥抱 AI？本文从政策、技术和商业三个关键词出发，一探 AI 赋能教育行业发展趋势。 政策利好，AI教育走向发展快车道 作为一个综合性概念，AI 赋能教育是指将人工智能技术深度融入教育核心场景和流程，促进关键教育场景的智能化以及关键业务流程的自动化。这种新 型教育模式，率先以「提效」二字闯入管理者及师生多方心坎，推动传统教育的教学主体、学习主体、教学载体开始变革。 具体表现是：基于精准学情分析，构建个性化教学体系；依托智能产品研发，拓展教育内容维度；借助大模型数据处理能力，实现规模化资源共享。教育 者们念兹在兹的不可能三角「个性化-高质量-大规模」，似乎突然有了解。 一切皆因果，因不虚 ...

Group 1: Technology Developments - OpenAI's CEO Sam Altman indicated that GPT-5 is likely to be released in the summer of this year, although delays may occur due to naming, safety testing, and feature iterations [2] - Analyst Ming-Chi Kuo reported that Apple's foldable iPhone is expected to begin production by the end of Q3 or early Q4 2025, with a total order of 15 to 20 million units [3] - MiniMax launched its video generation model Hailuo 02, which can produce 1080p original videos and has improved training and inference efficiency by 2.5 times [7] Group 2: Industry Trends - China Mobile's chairman Yang Jie stated that silicon-based life forms will outnumber humans in the future, becoming a significant part of the labor force and intellectual resources [4] - Huawei's rotating chairman Xu Zhijun highlighted four growth trends in the telecom market, including the emergence of new high-value user groups like streamers [5] - Amazon's CEO Andy Jassy mentioned that AI will lead to a reduction in workforce as the company implements more generative AI and intelligent agents [6] Group 3: Financial Updates - Aobi Zhongguang expects a net profit of approximately 55 million yuan for the first five months of 2025, with a revenue of around 363 million yuan, marking a year-on-year increase of 117.18% [9] - Li Yang Chip announced an initial pricing of 15.59 yuan per share for its inquiry transfer, with 15 institutional investors participating [13] - Texas Instruments plans to invest over $60 billion in expanding semiconductor manufacturing facilities in the U.S. [12] Group 4: Market Movements - JD.com officially announced its entry into the hotel and travel industry, offering hotel partners a three-year commission-free plan [8] - Samsung is reducing DDR4 supply, leading to a potential shortage that may last until Q3, with prices for DDR4 components rising significantly [16] - The "Chinese version of Nvidia," Moore Threads, has progressed in its IPO process, with a total of over 4.5 billion yuan raised in six funding rounds [10] Group 5: Corporate Actions - Intel plans to cut 15% to 20% of its chip manufacturing workforce, potentially affecting over 10,000 employees [11] - Huazhi Technology's subsidiary plans to invest 25 million yuan in a private equity fund focused on emerging industries in the Yangtze River Delta [18] - The joint venture between Rike Chemical and Ronghui Company aims to establish a technology firm with a registered capital of 40 million yuan [19]

Group 1 - Apple aims to leverage generative AI to accelerate the design of custom chips, which are crucial for its devices [1] - The comments were made by Apple's hardware technology senior vice president, Johny Srouji, during a speech in Belgium while receiving an award from Imec, a semiconductor research organization [1] - Srouji reviewed Apple's journey in developing custom chips, starting from the A4 chip for the iPhone in 2010 to the latest chips used in Mac computers and Vision Pro headsets [1] Group 2 - A key lesson learned by Apple is the necessity of using advanced tools for chip design, including the latest chip design software from Electronic Design Automation (EDA) companies [2] - The EDA industry is dominated by two major players, Cadence and Synopsys, who are competing to integrate AI technology into their products [2] - Srouji emphasized that generative AI has significant potential to complete more design work in a shorter time, greatly enhancing productivity [2] Group 3 - Another critical lesson from Apple's experience in self-developed chips is the importance of commitment, as demonstrated when Apple transitioned its Mac computers from Intel chips to its own without any contingency plans [2] - The shift to Apple’s custom chips for Macs was described as a gamble, with no backup plans or dual-chip strategies, requiring substantial software adaptation efforts [2]

每经记者｜蔡鼎 每经编辑｜兰素英 当地时间周三，Alphabet旗下的无人驾驶出租车（Robotaxi）公司Waymo宣布，下个月其出租车服务将 重返纽约市。之前在2021年，Waymo已在曼哈顿部分地区进行了初步的地图测绘和数据收集，此次回 归是为了进行更多的地图绘制以及测试工作。不过在初始阶段，Waymo不会提供完全的无人驾驶出租 车服务，而是将配备人类驾驶员，因为纽约州的法律不允许车辆在无人驾驶的情况下行驶。Waymo表 示，它正在推动修改法律，已经申请了纽约市的许可证，允许其出租车在曼哈顿自动驾驶，司机座位上 会有安全员。 点评：Waymo无人驾驶出租车重返纽约市进行地图绘制与测试，虽因当地法律限制需配备人类驾驶 员，但其推动法律修改并申请自动驾驶许可，展现了其在无人驾驶商业化落地上的持续推进与努力。 NO.4 OpenAI发布Altman专访，首次大谈GPT-5、AGI ｜2025年6月19日 星期四｜ NO.1 消息称特斯拉Grok车载AI助手即将上线：支持多个性定制 尽管特斯拉尚未将AI助手Grok集成到其车辆中，但相关固件代码显示，未来该功能将具备更逼真的交 互体验。马斯克曾在数月前透露，Gr ...

Core Insights - The total amount of venture capital investment in generative AI startups has reached $85 billion (approximately 610.3 billion RMB) from Q1 2022 to mid-June 2024, with a significant increase in average deal size to $372 million (approximately 2.67 billion RMB) [1][2][9] - Major investors such as SoftBank and Thrive Capital have emerged as leaders in this space, with SoftBank leading five rounds totaling $12 billion (approximately 86.2 billion RMB) and Thrive Capital leading eleven rounds totaling $8.9 billion (approximately 64 billion RMB) [5][8] - The investment landscape is becoming increasingly concentrated, with top firms dominating funding rounds, while smaller startups face challenges in securing financing [12][13] Investment Trends - The venture capital landscape for generative AI has seen a total of 724 rounds of funding, with the largest deals being concentrated among a few key players [1][2] - Notable investments include a $10 billion (approximately 72 billion RMB) round for OpenAI led by SoftBank and Thrive Capital, reflecting a trend of significant capital flowing into leading AI firms [10][11] - The average deal size has increased dramatically, with the top nine investors leading 74 rounds totaling $27.5 billion (approximately 197.4 billion RMB) in the past year alone [9] Investor Activity - Andreessen Horowitz has become the most active investor with 48 rounds led, while other firms like Accel and Lightspeed have also increased their investment activity significantly [5][9] - The ranking of venture capital firms is based on the total amount of capital they have led in funding rounds, which may overstate their risk exposure to individual companies [2][3] - The focus of investments is shifting towards companies developing leading models and commercial applications, with substantial funding directed at firms like Scale AI and ElevenLabs [10][11] Market Dynamics - The influx of capital into AI startups has led to a bifurcated market, where top-tier companies attract significant investment while smaller firms struggle to secure funding [12][13] - The recent acquisition of a stake in Scale AI by Meta for $14.3 billion (approximately 102.8 billion RMB) signals strong investor confidence in the potential of AI technologies [12] - There are concerns about a valuation bubble as top firms continue to receive funding, while smaller players face difficulties, leading to a potential risk of market correction if expectations are not met [13]