NVIDIA, Hyundai, Boston Dynamics Expand Physical AI Partnership

NVIDIA CEO Jensen Huang visited Seoul this week and held meetings with Hyundai Motor Group executive chair Chung Euisun, reporting participants and local press say (TechJuice; The Korea Herald). The parties described an expanded industrial roadmap focused on "physical AI" and robotics, with public coverage placing Boston Dynamics' humanoid Atlas at the centre of the collaboration (TechJuice; The Korea Herald). TechJuice reports Hyundai has indicated a target to mass-produce Atlas beginning in 2028, potentially at up to 30,000 units per year. The Korea Herald reports Hyundai has pledged 9 trillion won (about $5.9 billion) toward an innovation hub at Saemangeum that would host an AI data centre, a robot manufacturing cluster and a hydrogen plant. TechJuice reports the site is expected to run on tens of thousands of NVIDIA Blackwell GPUs. TechJuice also reports daily market reactions: Hyundai Motor shares rose nearly 7% and NVIDIA shares rose more than 6% on the news.

Per reporting in TechJuice, the parties described a development workflow that trains robots in simulation before deploying them into Hyundai manufacturing environments, with NVIDIA platforms supporting both simulation and the inference layer. The coverage emphasizes using large-scale simulation plus on-site inference and integration into existing industrial automation. TechJuice also describes a Seoul lobby converted into a live testbed where robots handled security, deliveries and plant watering during Huang's visit.

Companies integrating humanoid robotics with production-scale manufacturing typically rely on high-fidelity simulation to accelerate policy learning and to reduce physical wear during development. Observed patterns in comparable projects show simulation-to-reality gaps, robust perception stacks, and safety validation frameworks become central engineering efforts. For practitioners, that means large simulation clusters, long-running synthetic-data pipelines, domain-randomization strategies, and tightly instrumented field testing are likely priorities for teams executing similar programs.

The reported collaboration combines NVIDIA's AI compute and simulation tooling with Hyundai's manufacturing scale and Boston Dynamics' robotics IP, a mix public coverage frames as aiming to move humanoid robotics from demonstrations toward industrial deployment (The Korea Herald; TechJuice). If implemented at the scale described in reporting, the effort would be notable for combining industrial-scale production planning with edge inference deployments in factory settings, and for the implied demand for data-centre GPUs and robotics integration expertise.

• •Watch for formal announcements or memoranda from the companies or South Korean authorities that confirm site selection, financing, or timelines; The Korea Herald and TechJuice describe discussions but note no final decisions are publicly confirmed.
• •Track engineering disclosures about the simulation stack, safety validation, and how perception/control models are transferred from simulation to physical Atlas units; these will indicate how the project addresses sim-to-real risk.
• •Monitor procurement signals for large-scale Blackwell GPU deployments and partner ecosystems (cloud, middleware, systems integrators), which reporters say would underpin the Saemangeum hub (TechJuice; The Korea Herald).

Reporting describes discussions and figures attributed to outlets including TechJuice and The Korea Herald; Hyundai and NVIDIA have not issued a single consolidated public roadmap, and specific operational decisions (site selection, financing, and timelines) remain subject to confirmation in formal announcements.