硬·AI 作者 | 李 丹 编辑 | 硬 AI 特斯拉四季度营收由三季度的两位数同比大增转为下滑，电动车交付量连续第二年下滑，成为营收与汽车毛利承压的主要原因，但EPS盈利高于预期，能源与 服务业务继续放量，披露投资xAI、无人驾驶出租车Robotaxi等"物理 AI"相关业务的进展。 盘后公布财报后，股价盘后跳涨，盘后涨幅曾超过4%。盘后上涨更像是源于特斯拉短期的业绩并没有太差，长期的叙事——Robotaxi/Optimus/xAI方面继续给 估值提供想象空间。业绩电话会上，特斯拉高管称2026年全年资本支出将超过200亿美元。此后股价曾回吐盘后多数涨幅。 汽车业务拖累特斯拉四季度的总营收由增转降，但储能部署装机量创纪录，应被市场视为短期内对冲汽车端下行的"硬支撑"。更重要的是，特斯拉强调"物理 AI"的进展与AI领域相关投资。 Q4特斯拉营收同比由增转降3%、略低于预期、首次年度营收下滑，当季EPS盈利降17%仍强于预期，毛利率升破20%，储能部署增29%至新高14.2GWh；1月开始在奥斯汀限量提 供无安全监督员的Robotaxi；Cybercab、Semi卡车、储能系统Megapack 3均按计划今年 ...

Core Viewpoint - AsiaInfo Technology (01675) has seen a stock increase of over 5%, currently trading at 8.85 HKD, with a transaction volume of 66.69 million HKD, following the establishment of the "Embodied Intelligence Laboratory" in collaboration with ABB Robotics, marking a significant step in their strategic partnership in the Physical AI sector [1] Group 1: Strategic Developments - The "Embodied Intelligence Laboratory" was officially launched on January 25, symbolizing a substantial collaboration between AsiaInfo Technology and ABB Robotics in the Physical AI field [1] - Alibaba Cloud and NVIDIA participated in the ceremony, indicating their role in providing deep technical support for the laboratory's development [1] Group 2: Future Growth Strategy - AsiaInfo Technology has stated its commitment to an "AI-first" strategy by 2026, aiming to solidify its core operator service base and seize the strategic high ground in digital intelligence operations [1] - The company plans to cultivate a second growth curve through smart connectivity products, positioning itself as a leader in providing professional services, digital operations, and smart connectivity in the era of intelligent internet for enterprise clients [1] - The board and management express strong confidence in the company's future development prospects [1]

特斯拉称，这款新版Optimus将包含"相比2.5版本的重大升级，包括我们最新的手部设计"。公司正在为 第一条生产线做准备，该产线将于"2026年底前启动"。马斯克还预测，Optimus不仅将用于特斯拉工 厂，还将作为家庭助手甚至外科医生使用。 今天上午，人形机器人板块探底后，盘中迎来直线拉升，科大讯飞（002230）涨停，三维天地 （301159）等个股大涨。 当地时间周三盘后，特斯拉公布了好于预期的第四季度财报业绩，股价盘后上涨。值得一提的是，在财 报中，特斯拉表示，其面向量产的第三代Optimus人形机器人将于2026年第一季度亮相。 国内人形机器人行业最近也是催化不断，"我要上春晚"成为行业内近期的热词，魔法原子、银河通用、 宇树科技三家机器人公司接连官宣，与中央广播电视总台2026年春节联欢晚会达成合作。去年宇树机器 人在春晚扭秧歌出圈，引发市场极大关注，人形机器人板块走出一波大行情。 国信证券表示，马斯克对机器人产业进展的预期，表明行业的安全性和功能范围将在两年内实现极大程 度提升，机器人能力的提升也将带来需求端的爆发。持续看好人形机器人的长期投资机会，建议从价值 量和卡位上把握空间和确定性，从股 ...

机构：中邮证券 公司通过自研芯片化等大幅降本，为高毛利带来保障。硬件的降本来自设计，公司2017 年开始专注自 研芯片，2025 年宣布禾赛的高性能智能主控芯片费米C500，补齐了禾赛全栈自研的最后一块拼图。 公司成为目前行业唯一7大关键部件全栈自研的激光雷达公司。 核心看点：1）ADAS：L3有望成为最大变量。行业层面2025年政策、成本、技术齐发力，为高阶智驾 迎来拐点。L3牌照发放成为行业硬件需求结构性升级的核心变量。凭借ETX 主雷达+多颗FTX 补盲雷 达的产品组合，成功斩获首个乘用车量产定点项目，预计将于2026年底或2027 年初启动量产。2）机器 人：从25 年爆发的割草机器人开始，激光雷达不止于汽车，同时应用于泛机器人领域。我们认为，激 光雷达作为感知传感器，下游应用可以是任何需要"眼睛"去看的场景，激光雷达扮演的则是【卖铲子】 的角色、泛化能力强，长期市场空间巨大。 盈利预测 预计公司2025-2027年营业收入分别为30.9、41.0、54.1亿元，同比增加48.7%/32.8%/31.8%。GAAP 净利 润分别为4.2/6.8/9.2 亿元，25-27年同比增加508.5%、61.6 ...

虽然特斯拉四季度营收由三季度的两位数同比大增转为下滑，但储能部署装机量创纪录，应被市场视为短期内对冲汽车端下行的"硬支撑"。更重要的 是，特斯拉强调"物理 AI"进展与对外投资。 Q4特斯拉营收同比由增转降3%、略低于预期、首次年度营收下滑，当季EPS盈利降17%仍强于预期，毛利率升破20%，储能部署增29%至新高 14.2GWh；1月开始在奥斯汀限量提供无安全监督员的Robotaxi服务；Cybercab、Semi卡车、新储能系统Megapack 3均按计划将今年开始量产； Optimus第一代生产线正在铺设，目标今年底开始量产；预计Q1完成20亿美元认购xAI股票交易。特斯拉股价盘后一度涨超4%。 特斯拉四季度营收由三季度的两位数同比大增转为下滑，电动车交付量连续第二年下滑，成为营收与汽车毛利承压的主要原因，但EPS盈利高于预期， 能源与服务业务继续放量，披露投资xAI、无人驾驶出租车Robotaxi等"物理 AI"相关业务的进展。 盘后公布财报后，股价盘后跳涨，盘后涨幅曾超过4%。盘后上涨更像是源于特斯拉短期的业绩并没有太差，长期的叙事——Robotaxi/Optimus/xAI方 面继续给估值提供想象 ...

"2025年前三季度，智能眼镜市场出货量超过178万副，其中近八成是AI眼镜。"日前，工信部对外公布 的这组数据，无疑是"提振"了国内AI眼镜产业。如果按照规模来看，百万自然不是一个很大的量级，但 对比2024年20万左右的出货量，六倍的增长还是很可观的。 不仅如此，在2026年一开年，AI眼镜再一次刷屏，成为CES 2026上体验出圈的AI物种。资本层面，也 是好消息不断，包括雷鸟创新、影目科技以及XREAL等均有新的融资进账。随着字节、华为在今年也 都要下场，AI眼镜也将加速走进大众视野。刚刚上市的龙旗科技认为，智能眼镜板块预计将延续强劲 增长，2024年至2029年的复合年增长率达45.4%。微光科技副总裁、软工首席谢鑫更是直言，"今年市 场可能会达到千万级的销量。" 不过，仍需要指出的一点是，基于当前的硬件限制和软件生态的匮乏，AI眼镜远谈不上取代手机，最 终形态也存在争议，甚至长时间佩戴的问题也都没有解决。热潮之下，"百镜大战"会继续上演，不同的 是，今年更多巨头的入局会加速市场泡沫的破裂。而对于创业公司来说，则是一场没有退路的战役。 大厂押注、超10亿资金入账，新的卡位赛打响 当AI大模型在两年多 ...

Core Viewpoint - The article discusses the emerging trend of space computing power in the global AI competition, highlighting advancements from both American and Chinese companies in deploying AI models in space [1][4][13]. Group 1: Space Computing Power Developments - Starcloud, backed by Nvidia, has successfully run a large model in space, marking a significant milestone in space computing power [1][4]. - Guoxing Aerospace has announced the launch of the world's first silicon-based intelligent agent service network in space, planning to deploy 2,800 satellites to support billions of silicon-based intelligent agents [2][4]. - The total computing power from the planned satellites will reach 100,000 P-level for inference and 1,000,000 P-level for training, with full deployment expected by 2035 [4][6]. Group 2: Technological Differences - Starcloud's approach involves deploying large models on the ground before sending them to space, while Guoxing Aerospace can deploy general large models directly in orbit and update them as needed [9][10]. - This capability allows for real-time updates and operational flexibility, akin to over-the-air updates in smartphones [9][10]. Group 3: Advantages of Space Computing Power - Space computing power can significantly reduce costs and save land resources, as it operates without the constraints of terrestrial data centers [13]. - It offers energy efficiency by utilizing solar power directly in space, avoiding the high energy consumption associated with ground-based data centers [13]. - The real-time service capabilities of space computing power can enhance applications in various sectors, such as providing fishermen with timely information about fish movements [14][16]. Group 4: Challenges and Technical Considerations - The development of space computing power faces challenges such as hardware selection, the need for on-orbit hardware replacement mechanisms, and the unique environmental conditions of space [19][21]. - Issues like heat management and protection against high-energy particles must be addressed to ensure the reliability and accuracy of space-based computing systems [21][22]. Group 5: Future Outlook - The integration of space computing power with open-source large models presents a unique opportunity for China to establish a leading position in this emerging field [23][24]. - The ongoing advancements in both space computing and AI models are expected to drive significant changes in various industries, promoting broader access to AI technologies [17][24].

Core Insights - AI technology is experiencing three major breakthroughs: the evolution from chatbots to intelligent agents, the lowering of entry barriers through open-source models, and the understanding of the physical world through physical AI [1] - Physical AI is recognized as the next wave of AI development, showcasing its potential in understanding complex scientific principles [1] Group 1: VLA and World Models - The VLA (Vision-Language-Action) model and world models are emerging as a dual-model paradigm to address the data scarcity and safety issues in physical AI [2][3] - World models can generate infinite simulation data at a low cost, allowing VLA to learn from various scenarios without the risks associated with real-world data collection [3] - The integration of VLA and world models is seen as the optimal solution for enhancing embodied intelligence in physical AI [3] Group 2: Development Stages - The development of VLA and world models can be structured into four stages: cold start, interface alignment, training in simulated environments, and real-world transfer and calibration [4][5] - The cold start phase involves training a basic VLA model using existing robot datasets while the world model is pre-trained on vast amounts of video data [4] - The interface alignment phase focuses on mapping VLA's action outputs to the world model's input conditions to simulate the resulting scenarios [4] - In the training phase, VLA operates within the simulated environments generated by the world model, allowing for extensive reinforcement learning without physical wear on robotic components [4] Group 3: Addressing Challenges - Generative models often produce inconsistent outputs, leading to incorrect physical assumptions; introducing 3D geometry and material constraints can mitigate this issue [6] - A reward model can be implemented to evaluate the success of tasks in generated scenarios, providing feedback to the VLA [6] - The speed of world model predictions is crucial for training efficiency; techniques like latent consistency models can enhance prediction speed by focusing on feature changes rather than pixel-level details [6] Group 4: Data Sharing and Best Practices - The architecture of world models is evolving, but the necessity for real and synthetic data remains constant [7] - Sharing visual encoders between VLA and world models can optimize memory usage and ensure synchronized understanding of the environment [7] - Generating counterfactual data allows VLA to learn from hypothetical failure scenarios, improving robustness and reducing real-world testing costs [7] Group 5: Towards General Artificial Intelligence - The future of world models involves generating interactive 4D environments, enabling VLA to train in dynamic settings rather than static ones [8] - The integration of fast and slow systems within AI, where VLA handles real-time responses and world models manage long-term planning, is a key goal for advancements in autonomous systems [8] - Ultimately, VLA and world models may converge into a unified model capable of predicting both actions and future states, aligning with the vision of AI understanding physical laws [9][10]

Core Insights - WeRide has launched its self-developed simulation model, WeRide GENESIS, which bridges Physical AI and Generative AI, accelerating the development, training, and commercialization of autonomous vehicles [1][20]. Group 1: Technology and Innovation - WeRide GENESIS utilizes generative AI to quickly create realistic simulated urban environments, allowing for high-intensity training and validation of autonomous driving systems [3][4]. - The platform addresses challenges in the commercialization of autonomous driving by providing a comprehensive simulation solution that can replicate diverse city infrastructures and traffic behaviors [3][17]. - The simulation platform enhances the efficiency and safety of autonomous vehicle training by allowing AI drivers to experience a wide range of scenarios in a virtual environment [4][19]. Group 2: AI Modules - WeRide GENESIS incorporates four AI modules: AI Scenarios, AI Agents, AI Metrics, and AI Diagnosis, which collectively enhance the training and validation of autonomous driving algorithms [4][12]. - The AI Scenarios module simulates various critical situations that autonomous vehicles may encounter, ensuring comprehensive coverage of complex driving scenarios [8][9]. - The AI Agents module models the behavior of different traffic participants, enabling the simulation of realistic interactions and improving the robustness of decision-making algorithms [9][11]. Group 3: Performance Evaluation and Optimization - The AI Metrics module establishes a quantitative evaluation system that assesses driving behavior across safety, compliance, comfort, and efficiency dimensions, facilitating algorithm optimization [12][14]. - The AI Diagnosis module automates the identification of driving issues and provides actionable improvement suggestions, ensuring continuous enhancement of vehicle performance [14][16]. Group 4: Global Applicability - WeRide GENESIS is designed to be universally applicable, accommodating various urban road elements and vehicle configurations, thus streamlining the development process across different cities and vehicle types [17][19]. - The platform supports WeRide's position as a leading technology company with autonomous driving licenses in eight countries and deployments in over 40 cities worldwide [17][20].

Core Viewpoint - The GPGPU industry is transitioning from merely providing computational power to ensuring that the power is efficient, reliable, and cost-effective for real-world applications, especially in the context of AI and large models [1][3]. Group 1: Industry Trends - The demand for performance in AI has surged, with model training parameters growing from billions to trillions, necessitating a shift from simply increasing GPU numbers to addressing system engineering challenges [3]. - Data centers are evolving from hardware-centric operations to focusing on efficiency, reliability, and sustainability, with key metrics including PUE, TCO, and stability becoming critical [3][4]. - The average utilization rates for inference and training scenarios are low, highlighting inefficiencies in the current growth model of computational power [3][4]. Group 2: Company Developments - Tian Shu Zhi Xin has unveiled its fourth-generation architecture roadmap, aiming to surpass NVIDIA's Hopper architecture by 20% in performance by 2025 [6]. - The company is focusing on high-efficiency, predictable, and sustainable computing power, which is essential for long-term value [4][6]. - The introduction of innovative technologies such as TPC Broadcast, Instruction Co-Exec, and Dynamic Warp Scheduling aims to enhance performance and efficiency in their new architectures [8]. Group 3: Product Launches - The company plans to release multiple chip models, including the "Tian Gai" and "Zhi Kai" series, over the next three years, with a goal of doubling processing capabilities with each generation [9]. - The newly launched "Tong Yang" series includes various models designed for edge computing, emphasizing high performance and low latency for diverse applications [10][12]. - The Tong Yang series products have demonstrated superior performance compared to NVIDIA's AGX Orin in practical tests, showcasing their competitive edge in the market [12]. Group 4: Market Positioning - Tian Shu Zhi Xin aims to establish itself as a leader in the domestic edge computing market, focusing on high-performance, cost-effective solutions that connect AI with the physical world [12][20]. - The company has achieved significant performance improvements across various sectors, including internet AI, finance, and healthcare, with notable metrics such as a 70% increase in report generation efficiency [18]. - The firm emphasizes a comprehensive ecosystem approach, integrating software and hardware solutions to enhance user experience and performance [21][23].