Nvidia opens AI Day Seoul amid board probe into China-linked intelligence claims

Corporate crisis meets AI expansion

Nvidia's most ambitious event in South Korea, AI Day Seoul, began under a cloud of corporate scrutiny. The company's board announced an independent investigation into reports that CEO Jensen Huang was the target of a foreign intelligence operation allegedly linked to China. This probe was launched just hours before the conference sessions commenced at the COEX convention center.

The investigation stems from reports that a Singaporean-based executive at a data center company was acting as an intelligence asset for Beijing. This individual allegedly maintained a close personal relationship with Huang. While the board stated there is no current reason to believe US national security or Nvidia's confidential data was compromised, the timing of the probe creates a significant backdrop for the company's expansion in Asia.

Despite the internal investigation, the event proceeded with 20 partner companies and 15 technical sessions. Nvidia's spokesperson noted that Huang's interactions with customers are not unusual, as he meets with thousands of executives annually. The company remains focused on its technical roadmap while navigating these complex geopolitical and security allegations.

Accelerating chip and robot design

A major focus of the Seoul event is how Nvidia's software stack is being used to shorten industrial design cycles. Samsung Electronics demonstrated how they use the PhysicsNeMo AI framework to accelerate the modeling of semiconductor devices. This toolset allows for faster analysis of heat transfer and structural profiles, which previously took weeks on traditional computing infrastructure.

Similarly, SK Hynix is combining Nvidia's CUDA-X library with PhysicsNeMo to speed up simulation times for memory manufacturing. This partnership is structural, as SK Hynix is committed to co-developing next-generation memory for Nvidia's entire product line. These collaborations aim to compress the iteration loop for advanced node development in the global semiconductor market.

The event also highlighted advancements in physical AI, specifically for humanoid robots. LG Electronics showcased a training pipeline built on Nvidia's Isaac GR00T foundation model for robots. These technologies allow companies to simulate complex contact scenarios, such as robots grasping objects, in virtual environments before moving to physical trials.

Infrastructure and power constraints

The expansion of AI in South Korea faces significant hurdles regarding power and infrastructure. SK Telecom addressed these challenges by announcing a commitment to build a gigawatt-scale AI cloud using Nvidia's DSX framework. Their presentation focused on chip-level power delivery and grid constraints, with the first AI factory expected to go online in 2027.

To support these massive requirements, Naver Cloud is backed by a ₩10.7 trillion ($7.9 billion) syndicated loan. This is the first GPU financing of its kind in South Korea and is the country's largest-ever syndicated loan. The project involves a Brookfield-led initiative to acquire GPUs and supply computing capacity to Naver Cloud under long-term lease agreements.

Coupang also disclosed its infrastructure stack, featuring liquid-cooled racks built on Nvidia's GB300 NVL72 system. This liquid-cooling design is a direct response to the thermal density challenges that conventional air-cooled racks cannot handle. These developments show how Korean companies are engineering around the physical limitations of high-scale AI training and inference.

The strategic role of South Korea

South Korea has become a primary battleground for AI development due to its unique industrial base. The country is home to the world's two largest memory chip manufacturers and hosts a government-backed national AI project worth ₩2.08 trillion ($1.5 billion). This concentration of resources has led to a surge of events, with Intel and AMD also scheduling Korea-specific AI summits in November.

Jensen Huang has highlighted Korea's manufacturing expertise in semiconductors, automotive, and shipbuilding as ideal for robot deployment. These industries offer more physically demanding and sensor-rich processes than many other economies. This strategic alignment has led to expanded partnerships with major groups like Hyundai Motor, Naver, and SK Group.

The event also showcased the role of synthetic data in training AI models. Korean startups are using Nvidia's Cosmos 3 foundation model to generate training data for robots. By using synthetic data, these companies can replace the need for extensive and costly real-world data collection, accelerating the path to production-ready AI agents.

Safety and agentic AI deployment

As AI moves from experimental phases to production, ensuring the safety of autonomous agents is a growing priority. At the event, Korean company AIM Intelligence presented containment methods for these agents using Nvidia's NeMo Guardrails. These frameworks are designed to prevent agents from exceeding their intended permissions when granted access to enterprise systems.

Nvidia provided a hands-on program for attendees to build and run AI agents on-site using the NemoClaw framework. This allowed for live demonstrations of always-on agents running on Nvidia hardware. These tools are intended to help enterprises bound the actions of AI agents without eliminating their operational usefulness.

The event concluded with a focus on education and certification for the local workforce. Nvidia offered free certification exams to passholders covering AI model development and infrastructure. This initiative aims to build a skilled workforce capable of deploying the complex physical AI and safety frameworks showcased during the conference.