LG Is Building Its Humanoid on Nvidia's Stack, Not Its Own

LG and Nvidia signed a memorandum of understanding at Nvidia's Santa Clara headquarters on August 13, 2026, covering a next-generation bipedal humanoid, factory robotics, robot data infrastructure, and automotive computing. The humanoid gets a public unveiling in Q1 2027.

The interesting part is not that LG is building a humanoid. Nearly every large industrial conglomerate has announced one. The interesting part is the bill of materials.

LG brought the body. Nvidia brought everything above the neck.

The robot's foundation model is Isaac GR00T, Nvidia's open reasoning model for humanoids. Onboard compute is Jetson Thor. The safety framework is Halos for Robotics. That is the entire perception-to-action stack, sourced from one vendor.

What LG supplies is physical: actuators, sensors, and batteries drawn from LG Electronics, LG Innotek, and LG Energy Solution — the "One LG" capability set the company has been assembling for years across appliances, components, and cells.

This is a clean division, and it is worth being honest about what it implies. LG has decided that the differentiating asset in a humanoid is not the intelligence. It is the hardware supply chain and the manufacturing footprint. Everyone will have roughly the same brain; not everyone can build ten thousand precision actuators a month at appliance-industry cost.

That may be correct. It is also exactly the position Android handset makers occupied in 2010, and it did not end with the handset makers capturing the margin.

The Tennessee test is the part to watch

Before the humanoid appears, LG will test its CLOiD wheel-based robots inside an operating LG Electronics washing machine plant in Tennessee, later in 2026.

Wheeled robots in a working appliance factory is a far less glamorous announcement than a bipedal unveiling, and it is a far better signal. A humanoid demo is a controlled environment with a rehearsed task list. A washing machine line is a real production schedule with real takt time, where a robot that stalls costs measurable output.

Companies that put robots into their own factories first are running the honest version of the experiment. They eat their own downtime. If CLOiD works in Tennessee, the humanoid claims deserve more credit than the average unveiling video. If the Tennessee deployment quietly stops being mentioned, that is the answer too.

The data layer is the actual moat attempt

Underneath the robots sits PhysicalWorks, a data platform being developed by LG CNS, the group's IT services arm.

This is the piece that determines whether LG ends up with a business or a hardware line. Foundation models for robots are trained and refined on demonstration and telemetry data from real deployments. A conglomerate that runs appliance factories, HVAC installations, vehicle component plants, and a service technician network sits on top of an enormous amount of physical-world process data that no model company can buy.

If PhysicalWorks turns that into a training corpus LG controls, the GR00T dependency matters less — LG becomes the party with the proprietary post-training data, and Nvidia becomes the substrate. If PhysicalWorks stays a middleware project, LG is an integrator.

The compute commitments are the real money

Two dates in the agreement carry more capital than the robots do.

LG plans a reference AI factory built on Nvidia's Vera Rubin platform in the first half of 2027, and an 80-megawatt AI factory in Cheonan, South Korea, in the first half of 2028.

Eighty megawatts is a serious industrial facility — not a research cluster, not a demo. It is the scale at which a company is training and serving models continuously as a matter of operations. LG is committing to that capacity two years out, which tells you the humanoid is not a side project inside the group. It is the visible end of a much larger physical-AI bet.

The agreement also extends to Nvidia DRIVE Hyperion for next-generation vehicle computing, folded together with LG's infotainment and automotive software business. LG is one of the largest automotive component suppliers most consumers have never thought about, and pairing that catalog with Nvidia's autonomy platform is arguably the most immediately monetizable item in the entire MOU.

The quotes, and what they concede

LG's chief executive framed the deal as building "industry-leading references" to accelerate AI adoption. Nvidia's chief executive described the defining opportunity of physical AI as giving "every machine the ability to understand the real world."

Both statements are directionally true and strategically revealing. LG is selling references — proof that the technology works in a real plant, which it can then market to other manufacturers. Nvidia is selling every machine — the platform position, in which LG is one of many bodies attached to the same brain.

Neither party is confused about the arrangement. LG gets to skip five years of foundation-model work it would have lost anyway. Nvidia gets a tier-one manufacturer validating GR00T in production, plus an 80-megawatt customer.

The read

The specification list is more credible than most humanoid announcements because it is boring in the right places: named silicon, a named safety framework, a named factory, a named data platform, and dated milestones that can be checked.

What it does not contain is a target application, a production schedule, or a price. LG disclosed none of those, which means the Q1 2027 unveiling is a capability demonstration, not a product launch.

The company to compare LG against is not Figure or Tesla. It is Samsung, LG's mirror across every category it competes in, and the question is whether an appliance conglomerate's real advantage — supply chain, factories, and decades of servicing machines inside people's homes — converts into a robotics business, or just into being a very good contract manufacturer for someone else's platform.

The Tennessee plant will say more about that than the unveiling will.