# Physical AI in Korea: Samsung, LG and Kakao Mobility with Nvidia

> Physical AI puts models into factories, chips and cars. Nvidia's Korean work with Samsung, LG and Kakao Mobility shows what that looks like in practice.

- Published: 2026-11-09
- Topic: [AI](https://futuretechnologyhq.com/topic/ai/)
- URL: https://futuretechnologyhq.com/article/physical-ai-in-korea-samsung-lg-and-kakao-mobility-with-nvidia/
- Publisher: [Future Technology HQ](https://futuretechnologyhq.com/), free to read. Quote as "Future Technology HQ" with a link to the article.

## Key points

- Physical AI means models that perceive and act in the real world, trained heavily in simulation
- Samsung reports a roughly 1,600x speedup in one chip thermal simulation, from about 18 hours to 41 seconds
- LG and Kakao Mobility are testing robots and level 4 driving, but results so far are demonstrations and plans, not mass deployment

## Article

What does "physical AI" mean when companies such as Samsung, LG and Kakao Mobility use the term? Nvidia's latest Korea update gives three concrete answers: faster chip engineering, factory robots that learn from data, and a planned level 4 driving system built on Seoul's Gangnam experience.

## What physical AI actually is

Physical AI is artificial intelligence that has to deal with the physical world rather than a chat window. A language model that drafts an email can be wrong and be corrected a moment later. That is why the field leans so heavily on simulation, which acts as a flight simulator for machines: errors are cheap there and expensive everywhere else.

Nvidia's pitch is that Korea is well placed for this because of its strength in manufacturing and engineering. Its partners are applying accelerated computing (running heavy calculations on GPUs rather than conventional CPUs), AI and simulation to semiconductors, robotics and autonomous vehicles.

## How it works: the data flywheel and the speedup

The clearest mechanics come from LG Electronics. Nvidia and LG demonstrated Nvidia's Isaac GR00T model driving LG's CLOiD robot through manipulation and manufacturing tasks in an actual manufacturing environment. The collaboration implemented a physical AI data flywheel connecting these stages: real-world data is collected, augmented using Nvidia Cosmos, used to train GR00T, run on a Jetson AGX Thor onboard computer, then evaluated in simulation. The practical point is a repeatable process for training, evaluating and improving robots, rather than a one-off demo.

Samsung's example is more numerical. In chip-scale thermal and mechanical analysis, built with Nvidia PhysicsNeMo, Samsung combined three changes in one AI homogenisation case: AI-predicted effective material properties, coarser grids and GPU acceleration. The result was a roughly 1,600x speedup. A fine-grid solve on a 32-core CPU took about 18 hours; the GPU-accelerated coarse-grid version took 41 seconds.

## Who uses it and why

Samsung is using the approach across semiconductor engineering. In technology computer-aided design, it is working with Nvidia cuISS and cuDSS to accelerate sparse linear systems, a type of maths that underpins many simulations. Its sMOL materials platform has run selected quantum chemical calculations up to 5x faster end to end with Nvidia cuEST, and Samsung is exploring cuEquivariance to speed up tensor operations in atomistic AI models. The reason is plain: faster simulation means more design options tested before anything is fabricated.

Kakao Mobility is targeting a level 4 autonomous driving system, meaning a vehicle that drives itself within defined conditions without a human fallback. It plans to combine its own driving software and its experience operating in Seoul's Gangnam district with Nvidia Hyperion, a reference platform for vehicle sensors and compute, powered by DRIVE AGX Thor. It also plans to use Nvidia Halos for safety across development and validation. Initial work is planned around Kia purpose-built vehicles, with the aim of supporting future level 4 mobility services in Korea.

## What the limitations are

Read the verbs carefully. Kakao Mobility "plans" to build on Nvidia's technology; no service launch or date is given. LG's result is a successful demonstration in a manufacturing environment, not a fleet rollout. Samsung's 1,600x figure comes from one case, and it relies on coarser grids and AI-predicted material properties. The up-to-5x quantum chemistry gain is likewise for selected calculations. Samsung's idea of engineering agents that run analyses and compare design options, via Nvidia Agent Toolkit, is described as a future possibility.

There is also a dependency question. Every example runs on Nvidia's stack, from simulation libraries to onboard computers. That is convenient for the partners and a concentration risk for anyone who has to buy into it.

Power is another constraint. Separate reporting from Semiconductor Engineering notes that humanoid robots, autonomous vehicles and drones only work as long as the battery lasts, and that humanoid robots today may be useful for only a few hours per charge. Cadence's Frank Ferro argues that efficient processing, denser and faster memory, and optimised data movement all matter, and that no single component fixes it.

## What this means in practice

For engineering teams outside Korea, the useful lesson is the workflow rather than the brand. Simulation-first development, with real-world data feeding back into training, is being applied to manufacturing and driving problems, though so far through demonstrations and plans. Teams running CPU-bound simulation, such as thermal or materials modelling, have a published vendor-reported case to compare their own workloads against, bearing in mind that Samsung's figure depends on its specific setup.

UK and US firms in automotive, electronics and industrial robotics are the natural audience. The source names no UK deployment or pricing, so UK relevance is competitive rather than direct: these are the manufacturers and mobility operators that Western counterparts will be benchmarked against. Anyone attaching safety claims to autonomous systems should note that Kakao Mobility plans to use Nvidia Halos to support safety across development and validation.

## Where this is heading

The next steps are visible in the announcements themselves: initial level 4 development around Kia purpose-built vehicles, LG refining its robot training loop, and Samsung's engineering agents tying simulation tools together. The thing to watch is whether the demonstrations become production figures, such as robots working shifts at LG and driverless services running in Gangnam, and whether the power and memory constraints flagged by the chip industry ease enough to make humanoid robots useful for longer than a few hours.

For related reading on how AI is entering engineering tools, see [OpenAI and Synopsys are building a model that runs chip design tools](/article/gpt-synopsys-openai-chip-design-model/), and for Nvidia's work on agent controls, [Nvidia's AI agent kill switch runs on a separate chip](/article/nvidia-open-agent-safety-platform-openshell-sentry/).

## Sources

- [Nvidia blog, Korea advances physical AI across industries](https://blogs.nvidia.com/blog/korea-physical-ai-ecosystem/)
- [semiengineering.com](https://semiengineering.com/memory-in-physical-ai/)
- [businesskorea.co.kr](https://www.businesskorea.co.kr/news/articleView.html?idxno=280848)
