NVIDIA CEO Jensen Huang's Daughter Huang Minshan Attends Beijing Robot Conference to Explore Applications
NVIDIA senior executive Huang Minshan made a low-profile visit to the World Robot Conference in Beijing, where she observed humanoid robot demonstrations and inquired about sensor configurations at the companion robot booth.
Huang Minshan is the daughter of NVIDIA CEO Jensen Huang and is responsible for marketing NVIDIA's Omniverse and robotics platforms; the related software is used to simulate, test, and deploy "physical AI" systems, which are robots and automation devices capable of perception and action in the real world.
According to Reuters, she watched demonstrations of robots performing flying kicks and dances at the exhibition and stopped at the companion robot booth to learn about the sensors. Public reports confirmed her visit and on-site interactions but did not indicate that she represented NVIDIA in announcing purchases, investments, partnership agreements, or product orders.
NVIDIA's physical AI technology stack includes the Omniverse simulation library, Cosmos world foundation model, and Isaac robotics development platform. Its goal is to complete training, simulation, perception validation, and control testing using digital twins and synthetic data before physical robots enter real environments, thereby reducing the cost of trial and error in the real world.
The Beijing exhibition gathered Chinese robotics manufacturers, with a shift in focus from humanoid robot dance and performance to commercial viability, including industrial operations, service scenarios, and broader real-world deployments. Reports indicate that Chinese manufacturers are trying to prove to investors and customers that this technology is moving towards large-scale applications.
In market mechanisms, Chinese robotics manufacturers, component companies, and application customers are potential buyers of physical AI software, GPU computing power, and simulation tools; NVIDIA provides computing platforms, development frameworks, and simulation capabilities. If robotics companies expand training and deployment, capital will flow towards GPUs, edge computing, visual sensors, actuators, simulation software, and industrial integrators; those under pressure are robotics companies that rely solely on exhibition demonstrations and lack stable task data, customer orders, and deployment capabilities.
Source: Public Information
ABAB AI Insight
NVIDIA's entry into robotics did not begin with the humanoid robot craze. The company has long established a development ecosystem through CUDA, then entered edge AI computing with Jetson, and subsequently launched the Isaac platform to serve robot perception, navigation, manipulation, and simulation. Omniverse integrates industrial design, digital twins, and physical simulation into the same toolchain, while Cosmos further attempts to provide a world foundation model that can generate and understand physical environments. Huang Minshan is not examining a single hardware performance but whether these tools can create development and deployment demand within the Chinese robotics industry chain.
The core of the capital path is "selling shovels" rather than betting on a single robotics manufacturer. Robotics companies need to use data center GPUs for training, build synthetic environments for simulation, and use edge computing modules for deployment; NVIDIA can gain demand across four stages: model training, simulation, development, and inference. In contrast, Tesla bets on a closed loop of whole machines and manufacturing scenarios through Optimus, Figure AI relies on partnerships with companies like BMW to validate industrial tasks, and Apptronik seeks to expand deployment through industrial partners. NVIDIA's strategy is to provide a common technological foundation for multiple manufacturers.
A historical analogy is the Android and Qualcomm of the smartphone era. Google does not necessarily produce all phones but controls the application ecosystem through its operating system; Qualcomm does not need to create terminal brands but extracts value from industry shipments through chips and communication patents. If physical AI enters large-scale deployment, whole machine brands may become fragmented, but simulation environments, training frameworks, accelerated computing, and software toolchains will form a more centralized infrastructure layer. If Chinese manufacturers adopt NVIDIA technology, it will also improve their R&D efficiency but may simultaneously increase dependence on the American software and hardware ecosystem.
Essentially, this is a reconstruction of the industry chain. The current competition in humanoid robots is not just in mechanical structures but in the speed of the "data collection—simulation generation—model training—real deployment—feedback retraining" closed loop. Companies that have manufacturing scenarios and a large amount of real operational data can accelerate model iteration; suppliers with GPUs, simulation platforms, and developer ecosystems control training efficiency. The robotics value chain will shift from traditional hardware BOM cost competition to a composite competition of data, computation, world models, and deployment networks, with pricing power likely concentrating first on infrastructure and scenario entry points.
ABAB News · Cognitive Laws
Robots sell hardware, but compete on data closed loops
Those who occupy the scene first can train usable intelligence
The bottleneck of embodied intelligence is not in limbs but in the world model