Investment Rating - The industry investment rating is "Buy" with a previous rating of "Buy" as well [2]. Core Insights - The AI inference market is projected to reach a potential market size of $1 trillion by 2027, driven by the shift from training performance breakthroughs to optimizing inference efficiency [13][19]. - The demand for inference computing power is expected to grow structurally, with Deloitte predicting that the global inference workload will increase from approximately one-third of AI computing power in 2023 to about two-thirds by 2026, and potentially exceeding 80% in the long term [13][15]. - NVIDIA is transitioning from being a "shovel seller" in AI to an "AI factory," emphasizing the importance of integrated AI systems and infrastructure [19][30]. - The introduction of the NemoClaw platform aims to create a secure and controllable environment for AI agents, marking a shift towards "Agentic AI as a Service" [30][39]. - The physical AI sector is advancing with the release of the Cosmos world model and the Isaac simulation framework, which are expected to accelerate the deployment of robotics and autonomous driving technologies [41][46]. Summary by Sections Section 1: GTC Narrative Changes - AI inference chips are opening a potential market space of $1 trillion by 2027, with a focus on reducing inference costs and optimizing efficiency [13][19]. - The shift in AI narrative emphasizes the importance of inference capabilities over mere training performance [19][30]. Section 2: AI Applications - The introduction of NemoClaw enhances the OpenClaw ecosystem, providing a secure environment for AI agents [30][39]. - The physical AI domain is being bolstered by the Cosmos model, which integrates data generation and simulation for robotics [41][46]. Section 3: Investment Recommendations - The narrative in the AI industry is shifting from training-driven to inference-driven, with a focus on the cost-effectiveness and stability advantages of integrated AI solutions [47][48]. - The rise of Agent as a Service is expected to replace traditional enterprise SaaS products, indicating a collaborative evolution rather than a complete disruption of existing software [47][48].

Core Viewpoint - Geely and NVIDIA are set to collaborate on smart driving, smart cockpit, and smart manufacturing and research, focusing on three dimensions: physical AI, enterprise AI, and industrial AI [1] Group 1: Collaboration Focus - The collaboration will explore the continuous evolution of core capabilities in smart vehicles, leveraging NVIDIA's high-performance computing platform and Geely's deep expertise in vehicle-level scene understanding and multi-modal decision-making [1] - The partnership aims to advance physical AI technology architecture represented by the World Action Model (WAM), enhancing vehicles' environmental perception, behavior prediction, and collaborative execution capabilities [1]

Core Insights - Geely and NVIDIA are set to engage in deep collaboration focusing on smart driving, smart cockpit, and smart manufacturing and research [1] Group 1: Collaboration Areas - The partnership will concentrate on three dimensions: Physical AI, Enterprise AI, and Industrial AI [1] - In the Physical AI domain, both companies will explore the continuous evolution of core capabilities in smart vehicles [1] Group 2: Technological Framework - The collaboration will leverage NVIDIA's high-performance computing platform alongside Geely's expertise in vehicle-level scenario understanding and multi-modal fusion decision-making [1] - The aim is to advance the Physical AI technology architecture represented by the World Action Model (WAM), enhancing vehicles' environmental perception, behavior prediction, and collaborative execution capabilities [1]

Core Viewpoint - Geely and NVIDIA are set to engage in deep collaboration across various fields including intelligent driving, smart cockpit, and intelligent manufacturing and research, focusing on three dimensions: physical AI, enterprise AI, and industrial AI [1] Group 1: Collaboration Areas - In the area of physical AI, Geely and NVIDIA will explore the continuous evolution of core capabilities in smart vehicles, leveraging NVIDIA's high-performance computing platform and Geely's deep expertise in vehicle-level scene understanding and multi-modal fusion decision-making [1] - The collaboration aims to advance the physical AI technology architecture represented by the World Action Model (WAM), enhancing vehicles' environmental perception, behavioral prediction, and collaborative execution capabilities [1]

Core Insights - The article discusses NVIDIA's vision for the future of AI infrastructure, emphasizing the need for a complete redesign of the computing stack to support a multi-trillion-dollar smart economy [2] - NVIDIA aims to transition from a chip manufacturer to a comprehensive AI infrastructure provider, focusing on five layers: energy, chips, infrastructure, models, and applications [2] - The company predicts that global computing demand will exceed $1 trillion by 2027, with Token becoming the new foundational currency for technology companies [3] Group 1: Computing Demand and Infrastructure - NVIDIA's CEO Huang Renxun highlighted that by 2027, the global computing market will surpass $1 trillion, with the cost and efficiency of generating Tokens directly impacting tech companies' revenues [3] - The introduction of the Vera Rubin platform, which integrates CPU and GPU architectures, is expected to enhance computing capabilities significantly, allowing for the connection of up to 144 GPUs in a single system [5] - The new architecture is anticipated to optimize energy consumption and potentially deliver a revenue output ratio of up to 5 times for enterprises, reinforcing NVIDIA's dominance in the data center sector [5] Group 2: Ecosystem and Software Development - NVIDIA's CUDA ecosystem has reached a milestone with billions of GPUs installed globally, driving rapid advancements in AI technology and applications [6] - The launch of NemoClaw, a dedicated operating system for AI agents, allows developers to create personalized AI systems while ensuring privacy and security [7] - The company is also focusing on Physical AI, which requires AI to understand and interact with the physical world, as seen in partnerships with leading automotive companies for autonomous driving solutions [8] Group 3: AI and Industry Applications - NVIDIA is actively involved in various industries, including finance, healthcare, and manufacturing, by providing tailored AI solutions that enhance operational efficiency and decision-making [27] - The company is collaborating with major cloud service providers like IBM and Google Cloud to accelerate data processing capabilities, significantly improving speed and reducing costs [18][20] - NVIDIA's technology is being integrated into various platforms, enabling companies to leverage AI for real-time applications, such as supply chain management and customer service [28] Group 4: Future of AI and Token Economy - The article emphasizes the shift towards a Token-based economy, where the demand for computing power is expected to grow exponentially, driven by advancements in AI capabilities [31] - NVIDIA's infrastructure is designed to support this growth, with a focus on optimizing performance and cost-effectiveness across various deployment scenarios [34] - The company anticipates that the future of AI will involve a tiered pricing model for Token usage, reflecting the increasing value and demand for advanced AI services [46]

Core Insights - The GTC2026 conference highlighted NVIDIA's collaboration with global robotics leaders, including Zhiyuan Robotics, to advance the deployment of physical AI across various sectors such as manufacturing, logistics, transportation, healthcare, and infrastructure [1][3]. Group 1: Collaboration and Technology Development - NVIDIA's CEO Jensen Huang emphasized the deep partnership with Zhiyuan Robotics, showcasing their achievements in commercializing robotics for real-world applications and establishing a standardized closed-loop from R&D to deployment [3][6]. - The collaboration focuses on hardware platforms, edge-cloud computing, pre-training of the GR00T embodiment model, and industrial-scale production, aiming to define industry standards for physical AI [3][5]. Group 2: Product and Model Innovations - The introduction of the next-generation GR00T N2 embodiment model and the demonstration of the Spirit G1 as a hardware platform for model training were key highlights of Huang's presentation [3][5]. - Zhiyuan Robotics is working on joint hardware adaptations to support billion-parameter VLA models and world models for precise operations in electronic manufacturing, automotive assembly, and logistics [5]. Group 3: Standardization and Global Impact - Zhiyuan Robotics was recognized as a leading partner in global physical AI, establishing a value closed-loop with standardized capabilities across hardware, software, and deployment [6]. - The company aims to transform Chinese practical scenarios into internationally recognized standards, facilitating China's transition from technology follower to standard setter in the field of physical AI [6].

Investment Rating - The industry investment rating is "Outperform the Market," indicating that the industry index is expected to perform better than the benchmark index in the next 6-12 months [11]. Core Insights - Nvidia's CEO Jensen Huang presented at GTC 2026, highlighting the evolution of technology from Rubin to Feynman, with significant changes including the application of CPO and liquid cooling technologies, and the introduction of LPU with integrated large-capacity SRAM to enhance inference performance. Huang anticipates that agentic AI and physical AI will be key growth points for the future of AI [1][3]. - Nvidia's flagship computing platform, Vera Rubin, integrates seven types of chips and five types of racks, with expectations of revenue exceeding $1 trillion by 2027. The Feynman computing platform is expected to be released in 2028, further enhancing capabilities [3]. - The report emphasizes that CPO, liquid cooling, and LPU are foundational changes in AI computing infrastructure, focusing on performance and efficiency in inference applications [3]. Summary by Sections Investment Recommendations - Recommended companies to focus on include: - CPO Chips and Packaging: Tianfu Communication, Zhongji Xuchuang, Jingfang Technology, Changdian Technology, and Huanxu Electronics - Optical Fiber: Changfei Fiber, Hengtong Optic-Electric, and Zhongtian Technology - PCB and Materials: Shennan Circuit, Huitian Technology, Pengding Holdings, Shenghong Technology, and others - Server Assembly: Industrial Fulian - Power and Cooling: Yingweike, Magmi Tech, Lingyi Zhi Zao, and others - High-Speed Copper Cables: Zhaolong Interconnect and Wolong Nuclear Materials [3].

Core Insights - Huang Renxun emphasizes that the true vast industry lies in the integration of information and the real world, valued at $90 trillion [1][13] AI Development and Applications - Huang Renxun outlines a "five-layer cake" framework for AI, highlighting significant improvements in model performance and practical applications that create real economic value [2] - The emergence of AI agents has led to explosive growth in computational demand, validated by the rapid deployment of OpenClaw across various industries [2][3] - The next wave of AI development is identified as "physical AI," which aims to integrate AI with real-world applications in manufacturing and robotics, presenting substantial growth opportunities [3][4] Understanding the Real World - For AI to effectively understand the real world, it must grasp fundamental physical principles such as gravity, friction, and causality, which are often overlooked in traditional AI models [4] - Huang Renxun defines "physical AI" as an AI capable of understanding natural laws, moving beyond mere language processing to interact with the physical world [4][6] Strategic Partnerships and Innovations - NVIDIA has formed a significant strategic partnership with Dassault Systèmes to integrate AI capabilities with virtual twin technology, targeting industrial applications [9][10] - This collaboration aims to create an "industrial world model" that leverages scientific validation and physical principles for critical tasks in various fields [9][11] Impact on Manufacturing and Industry - The integration of AI in manufacturing is expected to revolutionize processes, allowing for extensive digital simulations and reducing reliance on physical prototypes [11][12] - Huang Renxun predicts that AI will enhance the efficiency of manufacturing processes, enabling even small and medium-sized enterprises to adopt advanced technologies [13] - The arrival of physical AI is likened to the transformative impact of the internet on information flow, suggesting a fundamental shift in how the physical world is designed and operated [13][14]

Core Insights - The AI industry has transitioned from the "generative AI" phase to the "inference inflection point" and the "Agentic era" as highlighted by NVIDIA's CEO Jensen Huang during the GTC 2026 conference [2] Group 1: Product Launches and Innovations - The GTC 2026 conference showcased the mass production of the Vera Rubin super AI platform, which is a vertically integrated system composed of seven chips designed specifically for Agentic AI, achieving a 35-fold increase in token generation capability per watt compared to the previous Hopper architecture [2] - The Vera Rubin platform's token generation rate in a 1GW data center increased from 2 million tokens per second to 700 million, marking a 350-fold improvement in performance [2] - NVIDIA has acquired the Groq chip team to integrate their technology into its product lineup, introducing the Groq LPU 30, which complements the Rubin GPU by excelling in low-latency decoding [2] Group 2: Strategic Directions and Market Predictions - Huang predicts that NVIDIA's revenue will reach $1 trillion by 2027, driven by increasing computational demands and the company's efficient token production capabilities [5] - The data center is evolving into a "token factory," where the number of tokens generated per watt directly influences revenue [5] Group 3: Open Source and Ecosystem Development - The open-source project "OpenClaw" was highlighted as a revolutionary development in the software ecosystem, with Huang stating it is outpacing Linux in adoption and is designed for AI agents [4] - NVIDIA introduced the enterprise-grade reference architecture NemoClaw to address security concerns associated with OpenClaw, ensuring safe operation of AI agents within enterprises [4] Group 4: Applications and Collaborations - NVIDIA is collaborating with companies like BYD, Hyundai, Nissan, Geely, and Uber to deploy a Robotaxi network, marking a significant step in autonomous driving [6] - In the robotics sector, partnerships with ABB, KUKA, and Disney are leveraging NVIDIA's technology for training humanoid robots [6] - The Vera Rubin Space One initiative aims to establish data centers in orbit, utilizing radiation cooling technology [6] Group 5: Industry Revolution and Future Outlook - Huang characterized the event as the "opening ceremony of the AI industrial revolution," emphasizing the ongoing transformation in the industry [7] - The conference featured interactive elements, including a performance by AI-generated music, symbolizing the integration of AI into various facets of life and industry [7]

Group 1: Nvidia Developments - Nvidia announced the launch of seven new chips based on the Vera Rubin architecture, marking the beginning of a new era in Agentic AI and establishing the largest AI factory globally [2] - CEO Jensen Huang stated that the demand for computing power is entering a phase of "million-fold growth," doubling the forecast for computing demand from $500 billion to $1 trillion by 2027 [2] - The next-generation Vera Rubin architecture features a fully liquid-cooled design and integrates Groq's deterministic flow processor technology, enhancing token generation speed by 350 times [2] Group 2: OpenClaw Project - OpenClaw is defined as the "Linux of the AI era," serving as an operating system-level framework that supports AI agents in autonomously calling tools and executing code [3] - The project aims to transition the software industry from SaaS to AaaS, with the introduction of the NeMo Claw reference design [3] Group 3: Alibaba's New Business Group - Alibaba established a new business group called Alibaba Token Hub (ATH), focusing on the creation, delivery, and application of tokens [4] - The new organization is directly overseen by Alibaba CEO Wu Yongming and includes several divisions such as Tongyi Lab and AI Innovation Division [4] Group 4: ByteDance's Seedance 2.0 Suspension - ByteDance has reportedly paused the global release of its latest video generation model, Seedance 2.0, due to copyright disputes with major Hollywood studios and streaming platforms [5] Group 5: Hello's Electric Bike Platform - Hello has removed all rental electric bikes from its platform following a compliance issue revealed during a consumer rights event [6] - The company has initiated an internal review due to the seriousness of the situation [6] Group 6: Zhipu's API Price Increase - Zhipu announced a 20% price increase for its new model API, marking the second price hike in recent times, with a total increase of 83% since Q1 2026 [7] - Following the announcement, Zhipu's stock price initially rose over 13% before settling at an 8.77% increase [7] Group 7: Funding News - Digua Robotics completed a $120 million Series B1 financing round, bringing total funding to $220 million across Series A and B rounds [8] - Blue Core Computing announced the completion of multiple rounds of financing totaling several hundred million RMB, with investments from various institutions [9] Group 8: New Product Launches - Zhipu launched its first "lobster model," GLM-5-Turbo, optimized for the OpenClaw scenario, showing significant performance improvements over previous models [10] - Apple's upcoming foldable iPhone is set to feature native split-screen functionality for the first time, enhancing user experience [11]