On June 8, Jensen Huang, CEO of NVIDIA, walked out of LG’s Twin Tower in Seoul with his arm around Koo Kwang-mo, Chairman of LG Group, and reportedly shouted “Go LG.” The gesture capped an hour-long meeting that formalized a sweeping partnership between the two companies: the joint construction of an AI Factory spanning robotics, cooling infrastructure, autonomous mobility, and sovereign AI. The announcement, made the same day, marks one of the most concrete outcomes of Huang’s Korea visit and signals how deeply NVIDIA is embedding itself into the hardware and data supply chains of the world’s largest industrial conglomerates.
Huang’s Korea trip began on June 5, immediately after his GTC Taipei keynote, where he confirmed the start of mass production for the Vera Rubin architecture and officially named Samsung Electronics, SK Hynix, and Micron as HBM4 suppliers. Within a week, those memory commitments were followed by collaborative agreements with multiple Korean business groups. The LG deal is the most multidimensional among them, involving no fewer than six LG subsidiaries across four distinct domains.
What Is an AI Factory and Why It Matters
What NVIDIA calls an “AI Factory” is not a rebranding of a GPU server cluster. The term describes a full-stack infrastructure platform that connects learning, simulation, verification, edge deployment, and digital twins into a single continuous workflow. The factory does not produce physical goods; it produces trained models, synthetic data, and validated AI behaviors at industrial scale.
This distinction matters because LG is positioning itself not merely as a company that uses AI, but as one that supplies the materials for AI production. The core of the collaboration is that LG Electronics will build a physical AI data factory dedicated to generating training data for robots. Using NVIDIA’s Cosmos world foundation model, the factory will produce and expand synthetic data, then supply it to companies in Korea and abroad developing robotic AI. LG’s decades of accumulated manufacturing data from home appliances, automotive components, and factory automation become the raw input for that pipeline.
The Four Collaboration Domains
The partnership is organized across four workstreams, each led by different LG affiliates with distinct responsibilities.
Robotics: Simulation, Training, and Hardware in One Loop
LG Electronics will integrate NVIDIA Isaac Sim and Isaac Lab into the development of its home robot brand CLOiD, enabling physically accurate simulation and reinforcement learning in virtual environments. The company is also evaluating adoption of NVIDIA Isaac GR00T inference models for real-time robot reasoning. Consumer robotics and household robots are the immediate target, with both companies committing to joint development in that category.
LG Innotek will contribute optical sensing components, while LG CNS will integrate the Isaac framework into its industrial and logistics robot platform called PhysicalWorks. The combined effort covers the full robotics stack: simulation, sensing, inference, and deployment.
AI Factory Infrastructure: Cooling, Modular Design, and Power
LG Electronics has already obtained NVIDIA certification for its cooling hardware, including coolant distribution units and cold plates. Now the company will move deeper into pre-hab modular design, building next-generation liquid-cooled AI factories aligned with NVIDIA’s DSX architecture. LG Uplus and LG CNS will deploy DSX-compliant AI factory infrastructure, while LG Energy Solution will supply 800VDC-compatible datacenter power solutions.
This workstream positions LG as a full-stack infrastructure provider for AI data centers, moving from component supply to integrated facility design.
Automotive and Mobility: ADAS, Cockpit, and Edge AI
LG Electronics will align its ADAS and in-cabin AI systems with the NVIDIA DRIVE Hyperion architecture. The company is also exploring next-generation cockpit controllers and edge AI processing built on NVIDIA DRIVE AGX. The automotive workstream extends beyond LG Electronics: LG Innotek’s sensing components will also feed into the same platform.
This is the most commercially immediate domain, given that LG Electronics already supplies automotive components to major global OEMs and can now offer NVIDIA-certified integration as a differentiator.
Sovereign AI: EXAONE on Blackwell
LG AI Research will strengthen its EXAONE large language model using NVIDIA Blackwell GPUs, the NeMo framework, and the Nemotron open dataset. This workstream builds on a strategic collaboration announced in April between LG AI Research and NVIDIA. The sovereign AI angle is explicit: LG aims to maintain independent control over its foundational model while relying on NVIDIA’s hardware and tools for scale.
The table below summarizes which LG affiliates are involved in each domain:
LG Group Affiliate Roles Across the Four Domains
- LG Electronics: Robotics, AI Factory Infrastructure, Automotive/Mobility
- LG Innotek: Robotics, Automotive/Mobility
- LG CNS: Robotics, AI Factory Infrastructure
- LG Uplus: AI Factory Infrastructure
- LG Energy Solution: AI Factory Infrastructure
- LG AI Research: Sovereign AI
The breadth is unusual. LG is the only partner in NVIDIA’s Korean strategy that touches robotics, datacenter cooling, automotive electronics, and foundation model development simultaneously.
Why This Partnership Is Strategically Critical for LG
LG Electronics has declared 2026 as the year its robotics business moves from pilot to scale. The company has already established a dedicated actuator brand, LG Actuator, and completed the acquisition of Bear Robotics, a self-driving robot company. These moves show hardware readiness. What has been missing is the training infrastructure to make robots operate reliably outside controlled environments.
The hardest part of robotics development is not the actuators or the compute board. It is the data pipeline: generating enough varied, physically accurate scenarios to train perception and manipulation models, then validating them without running millions of hours of real-world tests. NVIDIA’s simulation stack and Cosmos world model directly address that bottleneck. By coupling Isaac Sim with Cosmos, LG can generate synthetic training data at a fraction of the time and cost of real data collection.
Beyond its own robotics roadmap, LG sees a business opportunity in becoming a third-party physical AI data supplier. The same factory that trains LG’s robots can produce training datasets for other companies. This is a new revenue model for LG: selling data as a material for AI production, analogous to selling semiconductors as a material for computation. For a company that does not make its own AI chips, this is a logical and defensible position in the AI value chain.
NVIDIA’s Korea Strategy Extends Far Beyond Memory
The LG announcement should not be read in isolation. During the same Korea visit, NVIDIA secured HBM4 supply commitments from Samsung and SK Hynix, advanced discussions with Hyundai Motor on in-vehicle AI, and continued its existing collaboration with Naver on gigawatt-scale AI factories. The pattern is clear: NVIDIA is simultaneously locking in memory supply, manufacturing infrastructure, robotics data pipelines, automotive platforms, and sovereign AI capabilities across the Korean industrial ecosystem.
What makes LG different from other partners is the breadth of its portfolio. A single conglomerate supplies home appliances, cooling hardware, optical sensors, automotive components, telecommunications services, and energy storage. By partnering with LG as a group rather than with individual subsidiaries, NVIDIA gains access to a vertically integrated testbed for AI infrastructure that spans from the chip to the facility to the end device.
The timing also reflects a shift in NVIDIA’s own identity. The company is moving from selling AI chips to designing AI factories. LG, for its part, is moving from being a home appliance manufacturer to a robotics and AI infrastructure company. Those two trajectories intersected in Seoul on June 8.
What Comes Next
The partnership has been announced with executive visibility and cross-subsidiary commitment, but the real test is in execution. The robotics workstream must demonstrate that simulation-to-reality transfer works at scale for household robots. The cooling and modular datacenter workstream must deliver DSX-compliant facilities that match NVIDIA’s roadmap. The automotive workstream must produce production-ready ADAS integrations.
The sovereign AI track is perhaps the most closely watched. LG AI Research needs to show that EXAONE, trained on NVIDIA hardware, can compete with models built by companies with far larger compute budgets. Success there would validate the thesis that access to the right tools and data, not just raw compute scale, determines AI outcomes.
What is already clear is that the scope of this agreement far exceeds a standard supplier relationship. Jensen Huang did not travel to Seoul, spend an hour with LG’s chairman, and walk out arm-in-arm shouting “Go LG” to sign a routine procurement deal. The structure, breadth, and public visibility of the announcement suggest that both companies are betting on a long-term integration of their roadmaps. Whether that bet pays off will depend on whether the AI Factory concept can deliver what it promises: a complete, continuous pipeline from synthetic data generation to deployed robotic intelligence.
