Jensen Huang’s first stop in Korea wasn’t Samsung headquarters or a government ministry. It was a PC bang in Hongdae, where he sat down with League of Legends legend Faker. That choice wasn’t a publicity stunt—it was a deliberate signal about where NVIDIA came from and where it thinks it’s going.
Why Korea Mattered to NVIDIA Before Anyone Knew It Would
When NVIDIA was still a scrappy GPU startup in the late 1990s, Korean PC bangs were packing in gamers who demanded the sharpest graphics possible. StarCraft ran on NVIDIA hardware. Korean gamers, competing seriously and spending real money on performance, became some of the earliest mass-market customers for high-end consumer GPUs.
Huang acknowledged this directly during the visit: Korean gamers chose the best GPU available, and that was NVIDIA’s. The PC bang visit was a thank-you note written in symbolism—but it also set the frame for everything that followed over four days of back-to-back meetings with the heads of SK Hynix, Samsung, Hyundai, LG, Naver, Doosan, and others.
The Memory Problem Nobody Talks About Enough
A GPU is only as fast as the memory feeding it. For AI workloads—which require moving enormous volumes of data in and out of compute cores every second—standard memory creates a bottleneck. High Bandwidth Memory, or HBM, solves that problem by stacking memory dies vertically right next to the GPU die, slashing latency and multiplying throughput.
SK Hynix currently supplies an estimated 60–70% of the HBM4 going into NVIDIA’s latest Vera Rubin platform. Samsung’s vice chairman met with Huang to discuss next-generation HBM4E and HBM5 roadmaps. These aren’t peripheral supplier relationships—they’re load-bearing walls in NVIDIA’s AI infrastructure.
If NVIDIA can’t get the memory, the GPUs don’t ship. That dependency explains why Huang flew to Seoul personally rather than sending a procurement team.
Physical AI: The Phrase Worth Paying Attention To
Huang repeated one term throughout the Korea visit: physical AI. It refers to AI that operates in the real world—inside robots on a factory floor, in autonomous vehicles on public roads, in logistics systems moving physical goods. This is the frontier beyond chatbots and image generators.
The Korean partnerships he announced point squarely at this space:
- Hyundai — Discussions on autonomous driving and robotics collaboration, plus a positive review of a potential AI data center investment in the Saemangeum development zone.
- LG — Joint development of humanoid robots using NVIDIA’s Isaac GR00T ecosystem, combined with LG’s expertise in cooling systems for AI factory design.
- Doosan Robotics — Upgrading robot operating systems using NVIDIA’s platform, targeting an agentic robotics solution by 2027. Huang’s daughter reportedly visited Doosan’s research center before the official trip began, which suggests the interest was genuine, not ceremonial.
- Naver — Building an AI factory initiative around Naver’s Sejong data center, explicitly aimed at sovereign AI: Korean companies running Korean AI on Korean infrastructure without depending on foreign cloud providers.
What Korea’s Role Actually Is—and the Open Question
Put it all together and a clear division of labor emerges. SK Hynix and Samsung supply the memory that makes NVIDIA’s chips viable. Hyundai, LG, and Doosan manufacture and integrate the robots that run NVIDIA’s physical AI software. Naver builds and operates the data center infrastructure. NVIDIA sits at the center of the stack, owning the platform everyone else plugs into.
That’s a powerful position for all parties—but it also raises a real strategic question for Korean industry. Being an essential supplier or integration partner inside someone else’s ecosystem is valuable. Being the company that defines the ecosystem is something different.
The honest read is that Korea is currently positioned closer to the first role than the second. The companies meeting with Huang are excellent at manufacturing, memory, and infrastructure. The question is whether any of them—or Korean startups nobody’s watching yet—will build the AI services and applications that run on top of the hardware layer and capture the highest-margin value.
The Practical Takeaway
For anyone tracking AI supply chains or industrial robotics investment, Korea just became a more important story than it was six months ago. The HBM dependency alone makes SK Hynix a critical node in global AI buildout. The physical AI partnerships with Hyundai, LG, and Doosan are early bets on a market that doesn’t fully exist yet but is clearly where NVIDIA is pointing its roadmap.
Huang said before leaving that Korea has enormous opportunity in robotics and AI infrastructure. That’s true. The more interesting version of the same sentence is: the companies that move from building the pipes to owning what flows through them will define whether Korea ends up as a strategic partner or a strategic vendor.
