NVIDIA’s physical-AI stack is set to redefine four critical layers of industrial infrastructure: robotics, construction machinery, large-scale power generation, and the printed-circuit-board materials that support next-generation AI servers. This partnership with Doosan Group will push automated reasoning from controlled factory environments into dynamic real-world settings, including construction sites and power grids.
The collaboration encompasses multiple Doosan divisions: Doosan Robotics for intelligent robotic arms, Doosan Bobcat for autonomous compact equipment, Doosan Enerbility for gas turbines and small modular reactors, and Doosan Corporation Electro-Materials BG for high-performance copper-clad laminates. Together, the companies aim to integrate AI capabilities that can perceive, reason autonomously, and perform on-device inference—expanding beyond traditional robotics into broader industrial applications.
Doosan Robotics is focusing on integrating NVIDIA’s Isaac Sim, Isaac Lab, Cosmos foundation models, Newton physics engine, and Jetson Thor into its Agentic Robot OS. The goal is to transition from simulation to real-world deployment with minimal drift, enabling dual-arm and humanoid robots for tasks like depalletizing and sanding. These AI agents will rely on calibrated physics models that accurately map simulated environments to real-world conditions.
Doosan Bobcat is applying the same physical-AI framework to compact autonomous equipment used in construction, landscaping, and material handling. The emphasis here is on specialized world models that allow machines to perceive changing terrain, reason about obstacles, and adapt tasks autonomously—creating a reference ecosystem for small-scale autonomy.
On the power side, Doosan Enerbility will pair NVIDIA’s AI-factory platform (DSX) with its portfolio of gas turbines, steam turbines, hydrogen fuel cells, and small modular reactors. The collaboration is not just about increasing power output but optimizing generation equipment for the thermal and load profiles of accelerated computing clusters, including low-carbon solutions that can scale with data-center density.
Doosan Corporation Electro-Materials BG is supplying advanced copper-clad laminates (CCLs) designed to handle high-frequency signals and tighter thermal budgets. These CCLs will need to balance signal integrity with power dissipation as AI accelerators push beyond 800 watts per GPU, ensuring that motherboards can route multi-terabit bandwidth without thermal constraints.
Despite the promise, challenges remain. The physical-AI stack is still in its early stages, and production-ready use cases for dual-arm robots and compact autonomous equipment are not expected until 2025 at the earliest. Power solutions for AI factories will also require regulatory approvals that could extend timelines. However, if the collaboration adheres to its roadmap, AI reasoning could move from factory floors into construction sites and data-center basements within two years.
