LG and Nvidia Lock the Full Stack Behind Physical AI

LG Group and Nvidia signed a strategic business collaboration memorandum of understanding at Nvidia headquarters in Santa Clara, locking timelines for a bipedal humanoid robot in the first quarter of 2027, a factory test of LG’s wheeled CLOiD robot this year, and an 80 megawatt AI factory in Cheonan by the first half of 2028. Chairman Koo Kwang-mo and Nvidia founder and CEO Jensen Huang attended with top LG affiliate chiefs.

The deal turns June’s broad AI factory talk into a three-pillar program that runs through LG’s manufacturing data, components, cooling, batteries and industrial software. The robot is the visible piece. The deeper move is the stack behind it.

The MOU That Set Real Dates

Two months after the Seoul top-management meeting, the Santa Clara MOU spelled out concrete work in robotics, AI factories and mobility. Koo said the tasks where the companies can work together had become clearer in AI factory, physical AI and mobility, and that LG would accelerate adoption by building industry-leading reference cases.

Huang framed the opportunity around machines that understand the real world, reason and act safely beside people. Building on years of collaboration, he said the companies are combining LG’s product engineering and manufacturing with Nvidia technology to speed robots, AI factories and autonomous vehicles.

Nvidia’s own account posted a welcome for Koo and LG leaders as the firms enter “the next chapter” of collaboration across AI infrastructure, physical AI and robotics. That post drew tens of thousands of views within hours.

Three Pillars Run Through the Whole Group

The partnership is not a single product deal. It routes Nvidia platforms through multiple LG companies that already sell into homes, cars, factories and data centers.

That table is the second-order story. A humanoid announcement usually stops at the shell and the foundation model. Here the same conglomerate supplies the muscles, eyes, power pack, factory software, cooling loops and the Korean-language model stack that sits on top.

How the Bipedal Robot Gets Its Parts

LG said it will unveil a next-generation bipedal humanoid in Q1 2027. Onboard compute is Nvidia’s Jetson Thor. Development runs on the Isaac GR00T open humanoid platform and Nvidia’s Holoscan for Robotics safety stack (also called Halos for Robotics in LG materials).

• Actuators from LG Electronics serve as the robot’s muscles.
• Sensors and optics from LG Innotek serve as its eyes and spatial awareness.
• Batteries from LG Energy Solution supply onboard power.
• Simulation and training use Isaac Sim, Isaac Lab and Cosmos-generated synthetic data before any hardware hits a floor.

LG also plans joint reference robots inside the GR00T ecosystem. The company is building its own robot foundation model in parallel, using factory data from the CLOiD trials to improve it while still riding Nvidia’s open stack.

CLOiD Heads to a Tennessee Wash Line First

Before the biped appears, LG will put its wheeled CLOiD home robot on a washing-machine production line at its Tennessee factory this year. CLOiD debuted at CES 2026 as a household helper for indoor tasks. The factory run turns it into a real-world data collector for manufacturing AI.

Production-line hours generate the messy physical data that simulation alone cannot invent. LG intends to feed that experience back into both its own models and the shared Nvidia pipelines. The same pattern appears in the June technical plan: combine LG’s global manufacturing know-how with Isaac, Omniverse and Cosmos so the full path from raw materials to delivery can run on connected data and AI.

LG CNS is folding the same Isaac, Cosmos and GR00T pieces into PhysicalWorks so factories and logistics sites can adopt AI robots without starting from scratch.

From Cooling Parts to an 80 MW Cheonan Site

Nvidia’s June description of the NVIDIA and LG Group AI factory platform already listed cooling distribution units, cold plates and prefabricated modular design aligned with DSX. The August MOU turns that into site plans.

1. Prior baseline: LG Electronics pursued Nvidia validation for liquid-cooling gear, including a 600 kW CDU path, and treated AI data-center cooling as a growth pillar.
2. First half of 2027: LG builds a reference site on Nvidia’s Vera Rubin platform to prove its cooling, power and IT stack together.
3. First half of 2028: that expands into an 80 megawatt AI factory in Cheonan, South Chungcheong Province, using prefabricated design to cut construction time. The site will support physical AI work and LG’s robot foundation model.

LG Energy Solution is collaborating on 800 V direct-current power solutions for next-generation AI factories. LG Uplus and LG CNS plan scalable, power-efficient DSX-aligned facilities that can host the latest Nvidia GPUs and support GPU cloud services. The NVIDIA DSX AI factory reference design treats compute, networking, power, cooling and operations as one co-designed product aimed at lowest token cost, not a pile of separate racks.

That power-and-cooling layer connects to the wider pattern of Nvidia power investments for AI factories and the financialization of compute capacity. LG wants to sit on the facilities side of the same boom.

Mobility and EXAONE Ride the Same Rails

On vehicles, LG Electronics is aligning ADAS and in-vehicle AI with Nvidia DRIVE Hyperion and plans to use DRIVE AGX for AI cockpits and edge processing. The goal is to move beyond in-vehicle infotainment into a broader automotive electronics and software-defined vehicle platform. LG Innotek will tune sensing, connectivity and lighting modules for the same architecture.

Separately, Nvidia and LG AI Research continue work on EXAONE, Korea’s open sovereign model family. Training has used Blackwell GPUs, the NeMo framework and Nemotron datasets, with TensorRT-LLM for inference. LG is pushing EXAONE-based agents such as ChatEXAONE across the group. The June LG M.A.P. mobility AI physical plan already cast this as a national competitiveness play as much as a product roadmap.

Why the Stack Outranks Any Single Machine

Crowd reaction on X quickly zeroed in on the vertical integration: Jetson Thor plus GR00T paired with LG actuators, sensors and batteries, while the same group builds DSX factories. Physical AI starts to look less like a standalone robotics market and more like a new industrial computing stack that runs from synthetic data to the factory floor to the home.

The defining opportunity of physical AI is to give every machine the ability to understand the real world, reason and act safely alongside people, reshaping everyday life from the home and factory floor to the road.

Jensen Huang said that in the Santa Clara remarks reported by The Korea Times. The sentence fits the architecture Nvidia has been selling for months: simulation, foundation models, edge safety and factory-scale infrastructure as one product surface. LG supplies the missing Korean manufacturing density and the consumer-to-industrial product line that can actually ship the machines.

The same logic sits behind the broader shift of Nvidia AI compute as asset class. Once factories and robots become continuous consumers of tokens and simulation cycles, the partners that own cooling, power, sensors and deployment software capture a longer revenue stream than a one-time GPU order.

What Gets Built Before the Humanoid Walks Out

Between now and the Q1 2027 unveil, the concrete work is already scheduled. CLOiD must prove itself on a live U.S. production line. The Vera Rubin reference site must show that LG cooling and 800 V power pieces behave as a DSX-aligned system. LG CNS has to ship PhysicalWorks integrations that factories will actually buy. Mobility teams have to harden DRIVE Hyperion alignments into something OEMs can design around.

None of those steps is guaranteed. Prefab 80 MW builds slip. Safety certification for bipedal home and factory robots is slow. Sovereign model adoption inside large enterprises often stalls at pilot stage. The MOU removes ambiguity about intent and sequence; it does not remove execution risk.

What it does change is the shape of the bet. LG is no longer dabbling in robots on the side of appliances. It is wiring the group’s core strengths into Nvidia’s physical AI and AI factory platforms and publishing dates the market can track. For Nvidia, another full-stack industrial partner in Korea deepens the standard around Isaac, GR00T and DSX. For everyone watching physical AI, the next eighteen months will show whether a conglomerate that already makes the parts can also ship the integrated machines and the factories that train them.