Click the blue text and follow us
✦
JUMBO SHEEN
INVESTMENT
✦
Headlines

NVIDIA’s FY27 Q3 results exceeded market expectations, with its FY28 revenue growth guidance raised to 75%. Undelivered cloud orders stood at US$2.5 trillion, representing a quarter-on-quarter increase of US$500 billion. Vera Rubin NVL72 and Vera CPU have entered volume production and shipment. The board clarified that there is no plan for Jensen Huang to return. Following his departure, NVIDIA has reached out to three robotics startups in its investment portfolio and pledged continued support.
Huawei Cloud open-sourced CloudRobo-WM1.2, its Ascend-native embodied world model, which secured top rankings on both RoboTwin 2.0 and Libero Pro 2.0 benchmarks. Ascend Cloud now serves 80% of China’s embodied intelligence enterprises. The CloudRobo platform hosts 16,500 developers, over 230 enterprises and academic institutions, and more than 1,500 industry assets.
The MIIT approved 274 industry standards including AI – Key Fundamental Technologies – Technical Requirements for Embodied Intelligence Data Generation Systems. This marks the first time embodied data generation has been incorporated into the formal industry standard framework.
Guangdong’s 15th Five-Year Plan for Digital Guangdong sets out to strengthen high-end core robot components, build a provincial “1+1+N” embodied intelligence training ground system, and explore the development of world models. The Shanghai Stock Exchange voiced support for high-quality firms in future sectors such as embodied intelligence to list on the STAR Market.
The APEC CEO Summit (November 17–18, Shenzhen) served as a venue for concentrated industry insights. Wang Xingxing noted that the primary pain point remains insufficient model generalization, requiring retraining when switching scenarios. Zhou Jian projected that humanoid robots could reshape global industrial patterns over the next 20–30 years if widely deployed in flexible workflows. Lei Jun stated that intelligent robots hold greater potential than the automotive industry. Dong Mingzhu said Gree can supply reducers, drivers and servo motors.
Industrial robot output rose 26.0% year-on-year in October. In Beijing over the first ten months, service robot output surged 170% and industrial robot output grew 45.3%. Nevertheless, the secondary market saw sharp pullbacks for two consecutive trading days. On November 17, the CSI Robot Industry Index fell 4.0%, and Unitree dropped more than 10%.
SemiAnalysis slashed its 2028 AI capital expenditure forecast by nearly 30% and cut projected silicon wafer shipment area over five years by 17%, marking a notable divergence from NVIDIA’s robust guidance.
In-depth Analysis
Industry Updates: Huawei Cloud is targeting the "brain" segment for robot OEMs via open-source models with a clear platform strategy, and plans to roll out new models in Q1 next year. Siemens has partnered with Zhiyuan to demo agentic AI workflows: users input textual requirements, the AI automatically selects gripper types and adjusts geometric structures, followed by simulation validation and generation of additive manufacturing data. Industrial AI is extending toward embodied execution. Gree states that its CNC machine tools, motors, compressors, robot reducers, drives and servo motors are all domestically self-sufficient, positioning it as a cross-industry component supplier. General Motors aims to boost robotics and automation by 20% across its North American plants by 2028. Risen Smart plans a private placement of up to RMB 96 million for an intelligent logistics handling robot industrialization project (total investment: RMB 116 million, construction period: 18 months).
Technology & R&D Frontiers: CloudRobo-WM1.2 adopts PPA multimodal world representation, world-model-native DualAct action generation, cross-embodiment motion modelling and elastic inference. Trained on domestic Ascend computing resources, it has been deployed on multiple Ascend supernode clusters in China. Collaboration between Ascend and large models has therefore shifted from inference to training. CITIC Securities notes that OpenAI is restructuring a dedicated robotics team, and GPT-6 is expected to accelerate advances in embodied brains; Tesla may release footage of Optimus performing non-preprogrammed tasks this year. Microsoft Maia 200 has entered mass production, delivering up to a 1.7x improvement in tokens per watt, which helps cut embodied inference costs.
Academic Research: No major new peer-reviewed papers this cycle. Marginal changes include Huawei Cloud’s open-source model achieving top rankings on two public benchmarks, plus the State Administration for Market Regulation issuing the national standard guidance document Efficacy Indicators and Test Methods for Intelligent Computing Clusters. It unifies cluster efficacy evaluation for LLM training and inference for the first time, providing a common benchmark for university training grounds and computing infrastructure.
Policy & Macro Environment: Domestic policy follows a dual-track approach of “standards + scenarios”: technical requirements for embodied intelligence data generation systems have become industry standards, Guangdong has laid out core components and training ground frameworks, and the STAR Market is opening up to future industries such as embodied intelligence. Overseas, the Philadelphia Semiconductor Index tumbled more than 5% on November 16. SemiAnalysis cut forecasts for AI accelerator shipments and power demand, marking the first systemic downward revision to AI capex expectations this year. Humanoid robots and quadruped robots served as key demonstration exhibits at the APEC Summit in Shenzhen.
Capital Market Linkages: High-flying concept stocks in the robotics sector saw sharp profit taking (on November 17, over 100 stocks hit the 10% downside limit or fell more than 10%, including daily limit declines for Fuli New Materials and Jack Technology, and a drop of over 15% for Sunway Technology). This aligns with earlier clarifications from many firms that their robotics revenue share remains negligible. Under Hong Kong’s 18C regime, a wave of AI and robotics hard-tech firms has filed listing applications. Commercialization capability, order books and cash flow now separate valuations, with primary-secondary valuation inversion and post-listing underperformance common. The divergence between NVIDIA’s upbeat guidance and SemiAnalysis’ downgrade may amplify volatility across computing and embodied intelligence names.
Daily Investment Tips
1. The three-layer infrastructure of embodied intelligence covering models, computing power and data is being integrated by leading platforms (80% penetration of Ascend, open-sourced world models and rolled-out standards for data generation). Differentiation among robot OEMs will hinge more on scenario data accumulation and hardware reliability rather than self-developed model narratives. Validation should focus on genuine orders for training grounds, data services and domestic computing support.
2. Hardware players face dual pressure from in-house supply by robot OEMs and cross-industry expansion by home appliance and automotive component giants. This may squeeze the bargaining power of new entrants in the medium term. Validation metrics should be mass orders, yield rates and gross margins instead of capacity roadmaps.
3. A key risk lies in the disconnect between secondary market sentiment and primary valuations. The sector’s two-day sharp correction signals fading investor tolerance for pure concept plays. If upcoming catalysts (the V3 demonstration at Tesla’s shareholder meeting and mass delivery by leading robot OEMs) fail to meet expectations, compounded by downward revisions to overseas AI capital expenditure forecasts, valuation drawdowns may precede fundamental changes.
Jumbo Sheen FundDisclaimer:The content involved in this article is for sharing and exchange purposes only, and shall not be construed as investment advice. It is provided merely for readers' reference. All copyrights of the quoted content belong to the original authors. If any infringement is found, please contact us for removal.