NVIDIA and Doosan Group expand collaboration aimed at “physical AI” spanning robots, AI data center power, and key circuit materials

NVIDIA said it is expanding its collaboration with Doosan Group to pursue “physical AI,” an approach that links AI models and simulation tools to real-world machines such as industrial robots and construction equipment, while also addressing the infrastructure those systems need. The announcement, made June 7, 2026, links NVIDIA’s AI factory platforms and robotics software to multiple Doosan units including Doosan Robotics, Doosan Bobcat, Doosan Enerbility, and Doosan’s electronics materials business.

Under the expanded effort, NVIDIA says it will bring its full-stack accelerated computing platforms to Doosan’s work in industrial automation, power generation, and advanced electronics materials, with the goal of supporting next-generation AI infrastructure. NVIDIA also says Doosan’s businesses span several layers of the “AI factory ecosystem,” from intelligent robotics systems to large-scale power solutions and advanced materials used in data center equipment.

In the robotics track, Doosan Robotics plans to integrate NVIDIA Isaac Sim and Isaac Lab, NVIDIA Cosmos open world foundation models, the open source Newton physics engine, and NVIDIA Jetson Thor, into what Doosan calls its Agentic Robot OS. Isaac Sim is NVIDIA’s GPU-accelerated robotics simulation platform used to create and test robot behaviors in virtual environments, while Isaac Lab is a lighter-weight framework built for robot learning workflows and runs on top of Isaac Sim.

NVIDIA says Newton is an open-source, extensible physics engine built to support robot learning and development, and it is compatible with Isaac Sim and Isaac Lab. NVIDIA also positions Cosmos as a platform for open world foundation models aimed at physical AI development, and describes Jetson Thor as an embedded platform intended for physical AI and robotics workloads that can run local inference. NVIDIA’s June 7 update ties these components to Doosan Robotics’ goal of improving how industrial robots “perceive, reason and act” in complex and changing settings.

Doosan Robotics says it wants to use simulation-to-real workflows, physics calibration, and AI reasoning to make collaborative robots more adaptable and ready for scalable deployment. The company also says it is developing reference use cases for tasks such as depalletizing and sanding, and exploring new robot form factors including dual-arm and humanoid platforms. NVIDIA characterizes the effort as a step toward evolving Doosan Robotics from an arm provider into a broader, AI-first robotics solution company built around Agentic Robot OS.

On the infrastructure side, Doosan Enerbility says it is exploring opportunities to support AI factories, specifically the NVIDIA DSX AI factory platform, using its large-scale power infrastructure portfolio. NVIDIA describes DSX as an AI-factory platform designed to unify design, simulation, and operations to optimize performance and reduce costs across the infrastructure stack. In the Doosan collaboration, Enerbility points to generation assets including gas turbines, steam turbines, and small modular reactors, alongside Doosan Fuel Cell’s hydrogen fuel-cell systems, which it says are relevant to data centers needing reliable, high-efficiency, continuously available power.

The companies say future work could include power supply design for AI factory deployments, optimization of generation equipment, and evaluation of lower-carbon sources such as small modular reactors, with the aim of aligning AI infrastructure requirements with power-system expertise. Separate from compute and power, Doosan’s Electro-Materials business says it is supporting next-generation AI data center infrastructure through copper clad laminate (CCL), a material used in printed circuit boards (PCBs) that NVIDIA says are important for networking gear, AI accelerators, and AI server motherboards where low announcement loss and high reliability matter.

NVIDIA further links the materials effort to its MGX ecosystem, where MGX is a modular reference architecture that system makers and ecosystem partners can use to build accelerated servers and rack-scale AI infrastructure. In this collaboration, NVIDIA says MGX’s modular approach depends on system-level integration across compute and high-bandwidth networking, and that advanced PCB materials such as CCL may help enable high-speed announcement integrity as AI servers and related data center equipment increase in performance.

The companies did not provide timelines, performance benchmarks, or commercial terms in the June 7 announcement, and they did not say whether any of the robotics, power, or materials work would result in near-term deployments or specific customer sites. What to watch next is whether the collaboration moves from integration and reference use cases toward validated “simulation-to-deployment” workflows for robots, and whether DSX and MGX-related infrastructure conversations turn into measurable design guidance for AI data center buildouts that need both power reliability and high-speed data center announcement paths.