Nvidia held 'Nvidia AI Day Seoul 2026' at COEX in Seoul on Nov. 9 and announced a series of cooperation plans with the South Korean government and companies. It will expand the scope of sovereign AI cooperation with the Ministry of Science and ICT. It also teamed up with Rebellions and SK Telecom on AI infrastructure, with Samsung Electronics, Kakao Mobility and LG Electronics on physical AI, and with Hyundai Motor Group and others on AI agents.

◆Expanding government cooperation on sovereign AI... also holds startup partnership forum

Cooperation between Nvidia and the South Korean government is expanding from AI semiconductors and data centre infrastructure to talent development, startups, models, agents and security. It will first lay the groundwork for domestic neural processing unit (NPU) companies to join the Nvidia NVLink Fusion ecosystem. It will also run training and certification programmes and hackathons to build capabilities in using graphics processing units (GPUs). In models, it plans to support inference optimisation based on Nvidia Dynamo for 3 'AI for All' consortiums and 4 foundation model developers.

The two sides also held a startup partnership forum on the day. Nvidia announced an investment in AI interconnect company Point2 Technology. It said it also plans additional investments in Motif Technologies, Realworld and Upstage.

◆SK Telecom and Rebellions to build NVLink Fusion infrastructure

Rebellions will connect its next-generation NPU, 'Rebel R200', to the Nvidia platform using NVLink Fusion and develop systems based on Nvidia's MGX rack architecture. MGX is a structure in which GPU and NPU systems share rack space, power supply and cooling design. The resulting system will be deployed at SK Telecom's AI data centre and linked to its in-house model A.X K2 and the 'AI for All' public service.

The cooperation will be pursued with support from the Ministry of Science and ICT. It aims to support Korea-specific AI services and link system and operational capabilities needed for domestic semiconductor development and large-scale AI implementation. The three companies will also develop a distributed inference method in which Nvidia GPUs handle prefilling and Rebellions NPUs handle decoding. The R200 system is expected to be available for deployment from 2028.

Ryu Je-myung (류제명), the ministry's second vice minister, said the cooperation was meaningful because it created an important opportunity for domestically developed NPUs to enter the computing infrastructure market together with Nvidia, a global leader in AI computing, through NVLink Fusion.

Nvidia CEO Jensen Huang (젠슨 황) said NVLink Fusion would allow South Korean semiconductor companies to connect custom ASICs to Nvidia's AI factory platform.

◆Samsung Electronics, Kakao Mobility and LG Electronics cooperate on physical AI

Samsung Electronics is applying Nvidia's accelerated computing and AI physics across materials research, process development and device engineering. On its materials research platform sMOL, it used Nvidia cuEST to perform some quantum chemistry calculations up to 5 times faster. It also said it reduced simulations that took about 18 hours to 41 seconds in chip thermal and mechanical analysis based on PhysicsNeMo.

Kakao Mobility is developing a Level 4 autonomous driving system by combining Nvidia Hyperion and Halos with its own autonomous driving software and operating experience built in Gangnam, Seoul. Initial development will focus on Kia's purpose-built vehicles (PBVs).

LG Electronics demonstrated manufacturing tasks by its robot 'Cloi' based on Nvidia Isaac GR00T. In the process, the two companies built a workflow that links real-world data collection, Cosmos-based data augmentation, model training and simulation evaluation into a single chain.

◆Hyundai Motor Group and LG AI Research cooperate on agents and open models

Hyundai Motor Group is reviewing the adoption of enterprise AI agents using Nvidia's NemoCurator Blueprint and OpenShell Runtime. OpenShell is a runtime that manages an agent's data access and actions through sandbox execution and policy-based controls. Aim Intelligence is also integrating its security platform Starport with OpenShell to add functions for detecting malicious commands and data leaks.

In open models, LG AI Research will release an API for K-EXAONE-2.5 developed using Nvidia NeMoTron datasets. Separately, Nvidia released Dynamo deployment recipes for domestic open models including SK Telecom A.X K2, LG AI Research K-EXAONE 2.0, Upstage Solar Open 2 and Motif Technologies Motif-3.

Many of the announcements on the day were in planning and review stages, leaving tasks ahead. Hyundai Motor Group's agent adoption and Aim Intelligence's proof of concept (PoC) in the public energy sector are under review, and some startup investments are also at the planning stage. The operating model for national AI infrastructure was also discussed at the event. When Jung Gyoo Shin (신정규), chief executive of RABLab, asked how to integrate resource allocation, usage metering and cost management in national AI infrastructure that mixes multiple GPU generations and training and inference workloads, Nvidia said CUDA-based compatibility between GPUs has been secured to a significant extent. It said managing diverse generations of GPUs and workloads under a single system remains a task going forward.