NVIDIA and LG Build an AI Factory Across Six Subsidiaries

LG Group’s AI factory partnership with NVIDIA spans six of the conglomerate’s subsidiaries, linking home robotics, autonomous driving, data center cooling, energy storage, cloud infrastructure and a 236-billion-parameter sovereign AI model into a single integrated development workflow. LG Electronics shares hit the South Korean stock market’s 30% daily ceiling limit for two consecutive trading sessions this month, closing at 380,500 won (roughly $250 per share) on Monday, a quadrupling in value year-to-date after the deal’s scope became public.

LG Electronics is also building a physical AI data factory within the partnership, using NVIDIA Cosmos world foundation models to generate synthetic robotics training data it plans to sell to Korean and global companies pursuing physical AI and autonomous systems, with that pipeline as the deal’s commercially novel layer.

Six Companies, One Stack

Six LG subsidiaries each take a different slice of the NVIDIA technology stack, with roles that don’t overlap and stakes that are commercially new for a consumer electronics group.

• LG Electronics: home robotics (CLoiD home cobot), advanced driver-assistance systems (ADAS) aligned with NVIDIA DRIVE Hyperion, and the physical AI data factory
• LG AI Research: EXAONE sovereign AI model family, trained on NVIDIA Blackwell GPUs using the NeMo framework and TensorRT-LLM (a high-performance inference engine library) for deployment
• LG Innotek: next-generation sensing, connectivity and lighting components built to NVIDIA GPU architecture and optimized for robotics environments
• LG CNS: PhysicalWorks industrial robot platform integrated with NVIDIA Isaac frameworks, Cosmos world models and Isaac GR00T; building DSX-aligned AI factory infrastructure
• LG Uplus: a large-scale AI data center built to the NVIDIA DSX platform specification, in collaboration with LG Electronics and LG Energy Solution
• LG Energy Solution: 800 volt-direct-current data center energy solutions, validated against NVIDIA’s BESS (Battery Energy Storage System) Self-Qualification guidelines for next-generation GPU load profiles

Shawn Oh, head of Korea cash equities at NH Investment and Securities, said the collaboration “may broaden across LG affiliates” as the deal’s scope became clear. That read landed before the formal announcement did.

In March 2024, LG placed an initial $60 million investment in Bear Robotics, the U.S.-based startup building autonomous service robots for hospitality and food service environments. By January 2025, LG held a controlling 51% stake. The NVIDIA partnership connects those hardware bets to the simulation, training and deployment infrastructure they require to produce real commercial products.

The formal groundwork was laid in April when Madison Huang, NVIDIA’s senior director of product marketing for Omniverse and robotics and the eldest daughter of NVIDIA founder Jensen Huang, visited LG Electronics’ Seoul headquarters on April 28 to meet President Ryu Jae-cheol. The discussions covered cooperation in robotics, AI data centers and mobility.

LG’s Bid to Sell Robotics Training Data

Turning Manufacturing Know-How Into a Data Product

Robotics training data is expensive at scale. Real-world collection requires physical hardware, controlled environments and thousands of logged hours. Synthetic data generated in simulation is cheaper, but only useful when the physics are accurate. LG Electronics is building infrastructure designed to solve that equation for companies that cannot afford to build it themselves.

The official partnership announcement describes the division as a physical AI data factory that will “turn compute into data.” NVIDIA Cosmos world foundation models generate physically accurate virtual environments at scale. LG Electronics feeds its global manufacturing site data and production know-how into the pipeline alongside those synthetic sets, producing high-quality training datasets for robotics and industrial AI projects beyond its own product lines.

LG Electronics secured a 500 billion won backlog (roughly $380 million) within two years of setting up its dedicated smart factory business unit in 2024, per reporting from Digital Commerce 360. The data factory adds a distinct product line on top of that: AI training data generated by the same pipeline LG uses for its own robots, available to Korean startups without GPU clusters and global robotics companies that want region-specific training data from East Asian home and factory environments.

CLoiD and the Isaac Simulation Loop

CLoiD, the home cobot LG Electronics unveiled at CES 2026 in January, is the first product going through the full simulation pipeline. Built to handle indoor household tasks, it carries two articulated arms with seven degrees of freedom per arm and five individually actuated fingers per hand. LG is using NVIDIA Isaac Sim and Isaac Lab, the open robotics simulation and training frameworks, to validate CLoiD in physically accurate virtual environments before deploying it in homes.

Isaac GR00T, an open reasoning vision action language model built to give robots humanlike reasoning and the ability to plan and execute complex multi-step tasks, is also in the pipeline. NVIDIA and LG Electronics plan to jointly develop reference robots as part of the GR00T ecosystem, certifying LG hardware as a standard development platform within NVIDIA’s physical AI developer network.

LG CNS mirrors the logic at industrial scale. Its PhysicalWorks platform, integrated with Isaac, Cosmos and GR00T, targets logistics and manufacturing floors, running AI robot operations in simulation before any physical machine is commissioned. LG’s stated goal is an autonomous manufacturing ecosystem in which the entire process, from raw material procurement through production, logistics and customer delivery, runs connected in real time through data and AI.

EXAONE and the Sovereign AI Bet

South Korea ranks third globally in notable AI models, per the latest AI Index Report from Stanford University’s Human-Centered AI Institute, behind only the United States and China. Four of the five Korean models highlighted in that index come from LG AI Research’s EXAONE family: EXAONE Deep, EXAONE Path 2.0, EXAONE 4.0, and K-EXAONE.

K-EXAONE is the largest of the four, a 236-billion-parameter model that LG AI Research developed in roughly five months, unveiled at Korea’s first sovereign AI foundation model briefing hosted by MSIT (Ministry of Science and ICT). In internal benchmarking, K-EXAONE scored 72.03 on average, against 69.37 for Alibaba’s Qwen3 235B and 69.79 for OpenAI’s GPT-OSS 120B. The model runs on older NVIDIA A100 hardware as well as the latest Blackwell GPUs, lowering the entry cost for enterprises and smaller companies without access to current clusters.

NVIDIA’s role in EXAONE’s development is now explicit. LG AI Research used NVIDIA Blackwell GPUs for training, the NeMo framework for the training pipeline, Nemotron open datasets to improve training quality, and TensorRT-LLM for building high-performance inference engines. LG Group is distributing EXAONE internally through ChatEXAONE, its group-wide enterprise chatbot service built on the model.

MSIT is backing a broader sovereign model initiative that includes NAVER Cloud, NC AI, SK Telecom and Upstage alongside LG AI Research, all building Korean-language foundation models using NVIDIA NeMo and Nemotron datasets. The goal is AI infrastructure that operates independently of foreign platforms while meeting the performance standards of leading global systems. EXAONE Path, the healthcare model built with MONAI (a medical imaging framework for clinical AI), is already supporting cancer diagnosis research at startups and academic institutions.

LG Energy Solution’s 800-Volt Play

LG Energy Solution’s role in the deal reaches into infrastructure that standard AI partnership coverage rarely touches. The group’s battery and energy storage arm is working with NVIDIA on emerging 800V DC (800 volt-direct-current) data center energy solutions, aligned with NVIDIA’s BESS Self-Qualification guidelines. The 800V DC architecture targets power delivery at densities where standard voltage distribution loses efficiency: conversion losses climb as rack power draws cross 100 kilowatts per rack, a threshold next-generation GPU configurations routinely hit.

NVIDIA certifies LG Electronics’ cooling distribution units (CDUs) and cold plates for AI factory thermal management, the liquid-cooling components that carry heat away from dense GPU arrays running continuous workloads. LG Electronics is also developing prefabricated modular data center designs aligned with the NVIDIA DSX AI factory platform, compressing the gap between site selection and first active GPU workload. The company has already deployed NVIDIA Omniverse-based digital twins at nine production sites across five countries, including Korea, Mexico, Brazil, India and Vietnam, beginning at LG Smart Park in Changwon.

The DSX platform coordinates the full AI factory lifecycle through four software components:

• DSX MaxLPS: maximizes compute throughput and tokens generated per watt within a fixed power budget
• DSX Sim: validates AI factories as high-fidelity digital twins before and after physical construction
• DSX OS, an open-source operational software layer managing lifecycle, intelligent scheduling and resiliency across the facility stack
• DSX Flex: renewable and hybrid power orchestration, adjusting AI workloads dynamically in response to real-time grid signals

Cloud providers including CoreWeave, Crusoe, IREN and Lambda have already deployed DSX core components in live production environments. LG Uplus and LG CNS are the Korean industrial entrants to that ecosystem.

Korea’s GPU Buildout in Context

The NVIDIA-LG deal arrived after a wave of AI factory commitments from Korea’s major industrial groups, beginning at the APEC (Asia-Pacific Economic Cooperation) CEO Summit in Gyeongju, North Gyeongsang Province, in October 2025. The combined national commitment forms one of the largest AI infrastructure buildouts outside the United States.

Korea’s MSIT is separately deploying up to 50,000 of the latest NVIDIA GPUs through its National AI Computing Center, with an initial 13,000 Blackwell GPUs already operational via NAVER Cloud, NHN Cloud and Kakao Corp. NVIDIA’s total GPU commitment across Korea’s sovereign clouds and AI factories now exceeds 250,000 units, per the company’s October 2025 disclosure.

LG is the only major group in that table without a public GPU headcount. Its position is defined instead by the physical AI data factory scope and DSX alignment. If external demand for robotics training data materializes at scale, the GPU allocation follows the commercial order book.

The Stock Has Already Priced the Vision

LG Electronics largely sat out Korea’s AI rally this year. Samsung shares climbed 144% year-to-date as of early June on memory chip demand; SK Hynix gained 200% on the same thesis. LG watched both runs from the sidelines, classified as a consumer electronics and appliances company.

Reports that LG Group Chairman Koo Kwang-mo would meet Jensen Huang in Seoul, following Huang’s appearance at the GTC Taipei 2026 conference, changed that quickly. LG Electronics hit the 30% Korean daily ceiling limit for two consecutive trading sessions. LG CNS, LG Corp, LG Innotek and LG Uplus all rallied alongside it, each touching all-time highs on the same day.

The market repriced not who will build AI services, but who will construct the physical foundation to run AI.

Soowook Hwang, analyst at Meritz Securities, wrote that as the rally extended. Analysts at Kiwoom Securities and Hana Securities have tracked the same rotation: capital moving out of pure chip exposure, where Samsung and SK Hynix dominate, into companies that build or operate the physical AI stack, whether a robot arm, an ADAS sensor, a server cooling loop or a domestic-language enterprise model.

LG’s smart factory unit secured 500 billion won in backlog within two years of opening in 2024, and the company targets revenue in the trillions of won from that segment by 2030. The physical AI data factory is the newest line inside that business case. The NVIDIA partnership, through Isaac simulation, Cosmos synthetic data and DSX infrastructure, is the technical foundation LG intends to build it on.