Physical AI is no longer about individual robots performing isolated tasks; it's becoming a comprehensive industrial infrastructure that connects manufacturing, logistics, energy systems, and computing power into one coordinated ecosystem. Over the past week, major technology and manufacturing conglomerates have announced sweeping partnerships centered on NVIDIA's Physical AI platform, signaling a fundamental shift in how companies approach embodied artificial intelligence.

What's Driving the Shift From Robot-Centric to Infrastructure-Centric Physical AI?

The trend reflects a realization that standalone robots cannot reach their full potential without supporting infrastructure. Companies are now bundling Physical AI with AI factories, data centers, power solutions, and autonomous manufacturing systems. This integrated approach allows robots to learn from shared data, receive real-time computing support, and operate within optimized energy ecosystems.

LG and NVIDIA have launched an initiative called "M.A.P." that bundles Physical AI, AI infrastructure, and mobility into a single framework. Under this partnership, LG CNS is integrating NVIDIA's Isaac, Cosmos, and Isaac GR00T tools into its PhysicalWorks platform to accelerate AI robots for manufacturing and logistics. Meanwhile, LG Electronics handles cooling and modular AI infrastructure design, LG Energy Solution provides 800-volt direct current power solutions, and LG Uplus operates large-scale AI data centers supporting NVIDIA's Rubin GPUs.

Hyundai Motor Group is pursuing a similar strategy. During discussions with NVIDIA CEO Jensen Huang, the company outlined plans to expand collaboration from mobility and robotics into AI factories. Hyundai has separately announced a 9 trillion won investment (approximately $7 billion USD) with the South Korean government to develop an innovation hub in Saemangeum that will include an AI data center, robot manufacturing cluster, hydrogen production facility, solar infrastructure, and an AI-powered hydrogen smart city.

How Are Companies Building Physical AI Supply Chains?

• Memory and Computing Foundation: SK hynix and NVIDIA announced a multi-year technical partnership to co-develop next-generation memory for AI infrastructure, personal AI devices, and Physical AI systems. The collaboration includes memory for NVIDIA's Vera Rubin AI supercomputer, Vera CPU processors, RTX Spark-equipped personal computers, and Jetson Thor robotics platforms.
• Autonomous Manufacturing Operations: SK hynix and NVIDIA will accelerate semiconductor simulation and digital twin construction using NVIDIA's Omniverse platform and cuOpt optimization software, enabling autonomous operations of semiconductor fabrication plants.
• Agentic Robot Operating Systems: Doosan Group is integrating NVIDIA's Isaac Sim, Isaac Lab, Cosmos, Newton, and Jetson Thor to create an Agentic Robot OS that connects perception, reasoning, simulation, learning, and on-device inference. Doosan Robotics is developing high-value tasks such as depalletizing and polishing, plus new robot form factors including dual-arm and humanoid configurations.
• Equipment and Energy Solutions: Doosan Bobcat is exploring applications of NVIDIA's Physical AI technology to construction, landscaping, agriculture, and material handling equipment. Doosan Enerbility is investigating power solutions for AI factories using gas turbines, small modular reactors, and hydrogen fuel cells.

What Real-World Physical AI Demonstrations Are Happening Now?

Beyond corporate partnerships, tangible embodied AI systems are being deployed in real environments. At the International Conference on Robotics and Automation (ICRA) 2026, TARS showcased DexHand, a tactile robotic hand with 21 degrees of freedom that replicates human hand anatomy. The hand's fingertips contain ultra-compact cameras capable of capturing textures at 0.05-millimeter resolution at over 240 frames per second, allowing the system to understand physical properties like hardness, roughness, and slip risk.

TARS emphasized that DexHand's design prioritizes mass production capability, using only three types of motors and reducers in a quasi-direct drive configuration suitable for automated assembly lines. The system demonstrated real-time mirror control and performed 26 English alphabet sign language gestures, showcasing practical dexterity for real-world tasks.

In construction, GS Engineering and Construction partnered with Daedong Robotics to develop AI autonomous robots for construction site automation. The collaboration will conduct field trials of mass-produced AI robots at construction sites, starting with easily automatable tasks like material transport and repetitive work. GS E&C provides demonstration fields and site requirements, while Daedong handles design, research, and proof-of-concept work using its autonomous robot platform.

How Is Research Reshaping Physical AI Foundations?

Academic research is challenging conventional assumptions about Physical AI development. Recent papers on arXiv's robotics section question the prevailing focus on vision-language models (VLMs) and world models as the sole path to robot intelligence. Instead, researchers are proposing four critical interfaces: data interfaces that extract useful signals from human motion and internet videos; embodiment interfaces that map human movements to robot behavior; world model interfaces that perform physics-based 3D reasoning; and reward interfaces that estimate task progress from video and language.

A framework called PhyRoGen demonstrates practical applications of this research approach. It automatically generates combination puzzles with interdependent objects through procedural content generation, producing 24 physical puzzles from six types of generators. These puzzles can be solved in one to 300 seconds using sampling-based planning and are all manipulatable in KUKA LBR iiwa robot simulations, providing a benchmark for evaluating manipulation foundation models and synthetic data generation.

Why Should You Care About Physical AI Infrastructure?

The shift toward integrated Physical AI ecosystems has significant implications for manufacturing, logistics, construction, and energy sectors. Rather than deploying isolated robots, companies are building coordinated systems where robots, data centers, power infrastructure, and manufacturing facilities operate as a unified intelligence network. This approach promises faster deployment, better performance, and more efficient resource use. The partnerships announced suggest that Physical AI is transitioning from research demonstrations to industrial-scale implementation, with major corporations committing billions of dollars to infrastructure that will shape manufacturing and logistics for the next decade.