温馨提示 ： 点击下方图片，查看运营团队2025年最新原创报告（共210页） 说明： 欢迎约稿、刊例合作、行业人士交流 ， 行业交流记得先加入 "机器人头条"知识星球 ，后添加（ 微信号：lietou100w ） 微信； 若有侵权、改稿请联系编辑运营（微信：li_sir_2020）； 正文： 4月19日，备受瞩目的全球首场人形机器人半程马拉松在北京亦庄鸣枪开跑，20 支 由全国多家企业、高校和科研 机构组成的人形机器人队伍参赛。 大赛过后，最出圈的不是 冠军选手[北京人形机器人创新中心]的 全尺寸人形机器人— 天工Ultra，而是"清华 系"人形机器人创企[松延动力]，旗下人形机器人N2分别获得第二、第三名 （注： 松延动力有 两支赛队—"小顽童 队"和"旋风小子队"） ；至此，公司成为继宇树科技、众擎机器人后，2025年第三家"出圈"的人形机器人公司。 公司介绍： [ 松延动力 ] （Noetix Robotics） （公 司全称"松延动力（北京）科技有限公司"）于2023 年9月15日北京注册 成立， 研发方向涵盖通用人工智能本体、机器人仿生以及具身操作系统等。 公司成立至今，已发布了多款产品，且一直非常 ...

Core Viewpoint - The era of AI applications has arrived, marking a significant transformation in global business dynamics driven by AI super applications, which represent not just a technological revolution but a comprehensive evolution in business thinking [2]. Group 1: AI Super Applications - AI super applications are expected to evolve and bring disruptive changes to various industries, with a focus on identifying the next generation of super applications [3]. - The rise of DeepSeek illustrates a profound shift in public communication logic, highlighting the changing global landscape and the increasing sense of insecurity among populations, which amplifies any potential disruptive factors [5][6]. Group 2: Technological Trends - Embodied intelligence is a leading trend in AI research, merging physical interaction with autonomous decision-making, which is essential for the development of the next generation of general AI [6]. - AI is recognized as the most transformative technology in the last 50 years, reshaping industries and lifestyles, with super applications being empowered by ubiquitous AI engines [8]. Group 3: Industry Applications - AI applications are proliferating under the influence of large models, transitioning from single tools to professional services and from single agents to multi-intelligent collaborations across various fields [9]. - The education sector is experiencing a transformation with AI models that enable personalized, large-scale, and high-quality learning experiences, moving from a one-size-fits-all approach to tailored education [24]. Group 4: Entrepreneurial Opportunities - The current environment presents both challenges and opportunities for entrepreneurs, particularly in vertical fields where specialized model optimization and scene implementation can thrive [18][25]. - The importance of understanding the unique needs of specific industries and leveraging AI tools to create value is emphasized, as companies must focus on delivering tangible results [21]. Group 5: Data and Ecosystem - Data is identified as a critical element for future super applications, with companies focusing on building comprehensive data service platforms to drive innovation and integration with AI [41]. - The ecosystem for AI applications is evolving, with collaborative efforts among companies to enhance business efficiency and reduce costs through AI integration [45].

进入2025年，人工智能（AI）的狂飙突进令人瞩目。高效、便捷的人工智能，不仅是新质生产力发展 的重要力量，也极大冲击了各学科的研究范式。探讨与应用人工智能的新研究范式成为各学科的关注热 点。然而，尽管学术界抱有了巨大期望，尚处于大语言模型（LLM）的人工智能还未能达到通用人工 智能（AGI）的高度，AI起伏不定的幻觉特征与驳杂表现，也让一些研究者心存疑虑。模型上下文协议 （Model Context Protocol，MCP）作为AI领域的新技术，提供了一种沟通大语言模型与外部数据及工具 之间的联通机制，可以有效解决当前AI的"数据孤岛"，降低AI的幻觉几率，提升AI的智能化与自动化 程度。MCP协议推动AI更加接近AGI的发展目标，更对人机协同研究具有重要价值。 一、MCP协议的基本功能 大语言模型依赖静态的预训练数据，主要基于Transformer深度学习架构。该架构引入自注意力机制，用 于分析指令中不同序列的相互关系，并通过并行处理显著提升运算效率。简而言之，这种架构类似于翻 译过程，确保信息从源语言到目标语言的准确传递，使AI能够像人类一样"思考"，快速抓住关键信息， 并组织出最符合的语言表达。然而 ...

Investment Rating - The report maintains a "Strong Buy" investment rating for the company [2][6]. Core Insights - The company's performance met expectations, with stable gross margins and slight optimization in expense ratios. AI technology is driving stable growth in both B-end and C-end businesses, with the commercialization of the Spark large model accelerating [1][5]. - The company achieved a total revenue of 23.343 billion yuan in 2024, representing a year-over-year increase of 18.79%. The net profit attributable to shareholders was 560 million yuan, down 14.78% year-over-year, while the non-recurring net profit increased by 59.36% to 188 million yuan [5][7]. - The gross margin remained stable at 42.63%, while the net profit margin decreased by 0.95 percentage points to 2.17% due to increased bad debt provisions and reduced investment income [5][7]. - The company is significantly increasing its investment in the "Xunfei Spark" large model, with an additional R&D investment of 740 million yuan during the reporting period, establishing a solid foundation for competitive advantage in the AI industry [5][6]. Financial Data Summary - The company expects revenues of 27.673 billion yuan, 33.001 billion yuan, and 39.300 billion yuan for 2025, 2026, and 2027, respectively, with corresponding net profits of 738 million yuan, 1.072 billion yuan, and 1.312 billion yuan [6][7]. - The report indicates a projected PE ratio of 141.6, 97.4, and 79.6 for the years 2025, 2026, and 2027, respectively [6][7]. - The company reported a significant improvement in operating cash flow, reaching 2.495 billion yuan in 2024, a year-over-year increase of 613.40% [5][7].

Group 1: Government Policies and Initiatives - The Central Committee of the Communist Party of China and the State Council issued the "Opinions on Implementing the Strategy to Enhance Free Trade Pilot Zones," providing systematic deployment for the construction of free trade pilot zones [4] - The People's Bank of China, along with financial regulatory authorities, released the "Action Plan for Further Enhancing Cross-Border Financial Service Facilitation in Shanghai," which includes 18 key measures to improve cross-border settlement efficiency and optimize financial services [4] - The Ministry of Industry and Information Technology announced plans to promote orderly opening in the telecommunications sector and support pilot projects for service industry expansion, aiming to enhance market vitality [4] Group 2: Company News and Performance - Several companies reported significant year-on-year profit increases for Q1, including: - Bubu Gao: Net profit of 119 million yuan, up 488.44% [8] - Dajin Heavy Industry: Net profit of 231 million yuan, up 335.91% [8] - Yalian Machinery: Net profit of 57.6 million yuan, up 1088.67% [8] - Antong Holdings: Net profit of 241 million yuan, up 371.53% [8] - Guoda Special Materials: Net profit of 74.25 million yuan, up 1488.76% [8] - Xiantan Co.: Net profit of 48.02 million yuan, up 583.83% [8] - Shenghong Technology: Net profit of 921 million yuan, up 339.22% [8] - CATL announced the launch of its second-generation supercharging battery, capable of charging for 5 minutes to achieve a range of over 520 kilometers, marking a significant advancement in lithium iron phosphate battery technology [8] - China Communications Construction Company plans to repurchase A-shares worth 500 million to 1 billion yuan, while its controlling shareholder plans to increase H-shares by 250 million to 500 million yuan [8] - Rock Shares announced a termination of its major asset restructuring plan due to ongoing investigations by the China Securities Regulatory Commission [12] Group 3: Market Insights and Recommendations - The Shanghai Stock Exchange held a meeting with private equity representatives, who expressed confidence in the potential for valuation recovery in Chinese assets, suggesting a long-term, rational investment approach [5] - Citic Securities highlighted the importance of service consumption in expanding domestic demand, recommending investments in sectors like hotels and scenic spots [13] - Huatai Securities suggested focusing on defensive assets and policy hedging, particularly in transportation, insurance, and communication services, which are expected to see increased dividend rates in 2024 [13]

4月21日盘后，AI龙头科大讯飞（002230）（002230.SZ）发布2024年报，公司全年实现营业收入233.43 亿元，同比增长18.79%，同期归母净利润为5.6亿元。 公司核心赛道业务保持快速增长，消费者、教育、汽车、医疗业务营业收入分别同比增长27.58%、 29.94%、42.16%和28.18%，收入结构持续优化。 公司现金流创下历史新高，截至2024年末，公司全年经营性现金流净流入24.95亿元，同比增长超6倍。 2024年，科大讯飞研发投入达45.8亿元，占营收比例为19.62%。 分红方面，科大讯飞拟向全体股东按每10股派息1元（含税），预计将派发现金红利2.3亿元。 智慧教育业务实现收入72.29亿元，同比增长29.94%。投资者调研记录显示，2024年前三季度，科大讯 飞AI学习机销量增长超过100%。 此外，汽车、医疗、企业AI解决方案业务保持快速增长势头，分别实现收入9.89亿元、6.92亿元、6.43 亿元，分别同比增长42.16%、28.18%、122.56%。 值得一提的是，2024年12月30日，科大讯飞分拆医疗板块讯飞医疗（02506.HK）成功在港股上市，据 此前披 ...

Core Viewpoint - The future of AI technology is projected to evolve significantly, with robots potentially outnumbering humans in various sectors, including factories and households, as outlined by Zhang Yaqin, the director of Tsinghua University's Institute of Intelligent Industry Research (AIR) [1]. AI Technology Development Directions - AI large models are seen as a cornerstone of digitalization 3.0, with key development directions including multi-modal intelligence, autonomous intelligence, edge intelligence, physical intelligence, and biological intelligence [1][8]. - The transition from "digitalization 1.0" and "2.0" to "digitalization 3.0" involves a shift from small models to large models and from single-modal to multi-modal systems, indicating a broad application of AI across various industries [2]. Five Evolution Trends of AI Large Models - Large models and generative AI are expected to be the main technologies and industrial routes over the next decade, driving innovation and transformation [5]. - The ecosystem of AI will be significantly larger than that of personal computing and mobile internet, with foundational large models coexisting with vertical and edge models [6]. - Key elements of large models include tokenization and scaling laws, which enhance the model's ability to process diverse data types and improve performance with increased parameters and data [7]. Autonomous Intelligence - Autonomous intelligence will lead to personalized intelligent agents capable of self-planning, coding, and optimizing tasks, achieving high levels of autonomy and self-iteration [8]. - New algorithmic frameworks are necessary to overcome current inefficiencies and high energy consumption in existing algorithms, with potential breakthroughs expected in the next five years [9]. Path to General Artificial Intelligence - General artificial intelligence is anticipated to be realized within 15 to 20 years, with significant advancements expected in information intelligence, physical intelligence, and biological intelligence [10]. Future of Autonomous Driving - Autonomous driving is projected to be a key application of physical intelligence in the next five years, with safety levels expected to exceed human drivers by at least ten times [11]. - Large models and generative AI will enhance the generalization capabilities of Level 4 autonomous driving systems by generating high-quality edge case data and improving scenario simulation [12]. - The integration of multi-modal sensor data and end-to-end training will enable real-time collaboration between cloud-based large models and vehicle-specific models [13]. - Future autonomous driving applications will focus on single-vehicle intelligence, with a "vehicle-road-cloud" integration to ensure safety and optimize traffic flow [14]. - By 2025, autonomous driving may reach a pivotal moment, with 10% of new vehicles expected to have Level 4 capabilities by 2030 [15].

出品｜搜狐科技 作者｜杨锦 在大模型的热闹之外，朱松纯不惧做一个少数派。 他认为，科学本质是用简约的模型解释复杂的现象，比如杨振宁、爱因斯坦这些科学家，他们构建世界的模型里可能只有一两个参数。而今天的大模型，动 辄百亿甚至千亿级的参数，还不可解释，"从这个角度看，确实是非常丑陋的。" 但他也承认，从工程的角度来讲，大模型又能够在某些方面产生比较好的结果。 他和北京通用人工智能研究院（通研院）及北大几位老师最新合编的图书《通用人工智能标准、评级、测试与架构》，针对通用人工智能提出了系统的标 准、评级、测试与架构体系，对大模型也适用。 什么是智能？怎么样才算是通用了？在这套评测体系下，智能体不仅要完成任务，还需要自主定义任务，这意味着，智能体没办法像在其他评测体系上一 样"刷榜"。 过去20年，刷榜像附着在人工智能发展史上的藤壶，他本人也曾长时间地站在数据驱动和刷榜的一线。2004年，朱松纯和另一位全球计算机领域顶级科学家 沈向洋，在他的湖北家乡创建了莲花山研究院，是最早规模性地做大数据标注的机构。 刷榜是针对性地做端到端训练，而通用泛化的任务，是一个无穷的任务，比如训练机器人学会抓杯子，这很快，但问题是，稍微换 ...

Core Insights - The Volcano Engine has officially launched the Doubao 1.5 Deep Thinking Model, which utilizes the MoE architecture with a total parameter count of 200 billion and an active parameter count of 20 billion, achieving top-tier performance in multiple benchmark tests [1][3][8] Model Capabilities - Doubao 1.5 features practical capabilities such as "thinking while searching" and "visual understanding," available for enterprise users on the Volcano Ark platform [3][4] - The model can achieve a low latency of 20 milliseconds in high-concurrency scenarios, allowing it to perform searches and reasoning simultaneously [4][6] - It demonstrates visual understanding by analyzing text and image information, providing tailored recommendations based on user preferences [6][20] Performance Metrics - In various authoritative benchmark tests, Doubao 1.5's scores are comparable to OpenAI's models, particularly in mathematical tests like AIME 2024 and AIME 2025, while showing significant advantages in the ARC-AGI test [8][10] - The model scored 77.3 in the GPQA Diamond reasoning challenge, closely trailing OpenAI's models, and has shown strong performance in programming benchmarks [10] Market Position - As of March 2025, Doubao's daily token usage has exceeded 12.7 trillion, marking a threefold increase from December 2024 and a 106-fold increase from its initial launch [3] - Volcano Engine holds a 46.4% market share in China's public cloud model usage, positioning it as the market leader [3] Additional Model Upgrades - The upgraded Doubao Text-to-Image Model 3.0 can generate high-quality 2K images and is applicable in various fields such as marketing and design [11][15] - The new Doubao Visual Understanding Model enhances visual localization capabilities and supports semantic video search, making it suitable for commercial applications like security and home care [17][20] Industry Context - The competition among domestic reasoning models is intensifying, with Doubao 1.5's advancements in reasoning costs and visual understanding potentially setting the stage for the next wave of upgrades in the industry [21]

Core Insights - China Mobile has achieved annual revenue exceeding 1 trillion yuan and is advancing its cloud computing strategy for 2025, focusing on a comprehensive computing network that integrates various types of computing power [1] - The chairman of China Mobile, Yang Jie, highlighted two significant changes driven by innovations in general artificial intelligence: exponential growth in total computing power and a shift in computing power structure, with inference computing demand expected to surpass training computing demand in the coming years [1][2] Group 1: Computing Power Network - China Mobile has established the first nationwide computing power network that integrates general computing power (8.5 EFLOPS), intelligent computing power (43 EFLOPS), quantum computing power (1138 Qubit), and supercomputing power (800 PFLOPS), accounting for one-sixth of the national total [1] - The intelligent computing power scale is projected to grow over 2.5 times in the next three years, with an annual compound growth rate of nearly 40% [1] Group 2: AI Development Effects - Yang Jie identified two scale effects in AI development: a technological capability scale effect leading to a tenfold increase in AI capabilities annually, and an economic benefit scale effect that will reduce AI usage costs by approximately tenfold each year [2] - By 2030, AI tasks are expected to account for 64% of total network traffic, necessitating urgent innovations in computing network infrastructure [2] Group 3: Next-Generation Cloud Computing Paradigm - China Mobile released the "Cloud Intelligence Computing Technology White Paper," proposing a next-generation cloud computing paradigm that integrates cloud and AI [3] - The new architecture will evolve from traditional IaaS, PaaS, and SaaS to AI IaaS, AI PaaS, MaaS (Model as a Service), and AI SaaS, facilitating a transition from "computing network brain" to "intelligent computing network brain" [3][4] Group 4: Strategic Initiatives - China Mobile announced four strategic initiatives: building an integrated computing network infrastructure (AI IaaS), creating a self-service tool platform (AI PaaS), developing a one-stop model service (MaaS), and launching native product applications (AI SaaS) [6] - These initiatives aim to enhance AI innovation efficiency and support the comprehensive digital transformation of production, lifestyle, and social governance [4][6] Group 5: Emerging Business Models - The new strategy indicates the emergence of new consumption patterns in the cloud industry, particularly through the integration of AI smart terminals with cloud computing [7] - The launch of the mobile AI cloud computer, which integrates advanced GPU technology, aims to enhance user experience by significantly improving performance metrics such as CPU load reduction and graphics rendering capabilities [8] Group 6: Industry Collaborations and Future Focus - China Mobile, in collaboration with companies like Moore Threads, initiated the "Intelligent Computing Interconnection OISA Industry Chain Initiative" to enhance high-performance GPU interconnection technology [9] - The focus on smart connected vehicles is expected to grow, with projections indicating that by 2030, over 80% of new cars will be smart connected, supported by high-performance computing and low-latency networks [9]