NVIDIA says Samsung, Kakao Mobility and LG build on its physical AI stack

NVIDIA has published a post describing work with three Korean companies, Samsung Electronics, Kakao Mobility and LG Electronics, which it says is moving physical AI out of development and into real-world production. The performance figures come from NVIDIA alone and have not been independently verified. Earlier public statements from two of the partners add context.

What does NVIDIA say Samsung achieved?

NVIDIA says Samsung is using accelerated computing and AI physics across materials research, process development and device engineering. According to the post, Samsung's sMOL platform ran some quantum chemical calculations as much as 5x quicker overall with NVIDIA cuEST. For chip-scale thermal and mechanical analysis, Samsung is working with NVIDIA PhysicsNeMo. In one AI homogenization case, pairing AI-predicted material properties with coarser grids and GPU acceleration gave a roughly 1,600x speedup, cutting simulation time from about 18 hours on a 32-core CPU to 41 seconds on a GPU.

The 1,600x figure compares a fine-grid CPU solve with a coarse-grid GPU solve that uses AI-predicted material properties, so it reflects a change of method as well as of hardware. NVIDIA describes it as one case, not an average across Samsung's workloads.

How does the Kakao Mobility plan compare with its own roadmap?

NVIDIA says Kakao Mobility plans to use its Hyperion reference platform, powered by DRIVE AGX Thor, and its Halos safety technology to develop a level 4 system, starting with Kia purpose-built vehicles and pairing NVIDIA's vehicle computing platform with Kakao's proprietary driving software and its operating experience in Seoul's Gangnam district. These are plans, not a deployed service.

In April 2026, Kakao Mobility presented a level 4 roadmap at the 2026 World IT Show in Seoul. According to AI News, which reported the event, the company said it would develop level 4 technologies in-house, while planning to share datasets, HD maps and ride-hailing APIs with outside companies. The same report said Kakao's late-night Gangnam robotaxi, available through Kakao T, recorded 7,754 rides between September 26, 2024 and February 28, 2026, citing the Seoul Metropolitan Government, and became a paid service in April 2026 after a free pilot. We did not find a Kakao Mobility statement on the NVIDIA plans; the two accounts are not necessarily in conflict, but only NVIDIA has described this one.

What is the LG context?

NVIDIA says it and LG demonstrated Isaac GR00T-powered manipulation and manufacturing tasks with LG's CLOiD robot in an actual manufacturing environment, using a data flywheel that connects real-world data collection, Cosmos-based data augmentation, GR00T training, Jetson AGX Thor computing and simulation-based evaluation. The post does not name the site.

In August 2026, The Korea Times reported that LG Group and NVIDIA signed a memorandum of understanding under which LG hopes to unveil a bipedal humanoid built on Isaac GR00T in the first quarter of next year, and that CLOiD, shown at CES 2026, would be tested on a washing machine production line at LG's Tennessee factory. The same report said LG plans an 80-megawatt data center in Cheonan in the first half of 2028. Whether the demonstration NVIDIA now describes is that Tennessee test is not stated in either account.

What should readers take from it?

The post is a vendor's account of customer work, so the numbers are best read as NVIDIA's claims. It is consistent with a wider pattern: Hyundai Motor Group and Kia announced in March 2026 that Hyundai plans to build NVIDIA's autonomous driving technology of level 2 and higher into selected vehicle models, on an architecture that scales from level 2 to level 4, subject to a separate definitive agreement, and several Korean companies are working with NVIDIA's platform on robots and vehicles. Independent measurements of the Samsung speedup, and statements from Kakao Mobility and LG about the specific work described, would show how much of this has reached production.