Jensen Huang's Seoul Blitz: NVIDIA Locks In Six Mega-Deals as South Korea Bets on AI Sovereignty

Jensen Huang doesn’t do quiet visits. When NVIDIA’s CEO landed in Seoul on June 4 for a four-day trip, South Korea’s tech establishment mobilized around him — and by the time his plane left, he had signed partnership agreements with six of the country’s most important corporations. The deals span virtually every layer of the AI stack, from the memory chips that power GPU training clusters to the humanoid robots that may eventually work on assembly lines. They also tell a story about something larger: the scramble by non-US nations to lock in AI infrastructure before the window closes.

The Memory Deal That Underpins Everything

The most strategically significant agreement is the least glamorous: a multi-year extension of NVIDIA’s partnership with SK Hynix, the world’s second-largest memory chip manufacturer and the dominant supplier of High Bandwidth Memory (HBM), the specialized chip architecture that allows GPUs to move data fast enough to train and run large AI models.

The numbers are staggering. “We already buy from SK Hynix billions and billions of dollars each year,” Huang said at the announcement, “and it’s going to grow substantially.” The deal covers four computing platforms and positions SK Hynix as NVIDIA’s primary memory partner for the foreseeable future. Huang was explicit about the strategic relationship: “SK Hynix is NVIDIA’s largest memory supplier and will remain its largest partner in that regard as well.”

This matters because memory has quietly become the binding constraint in AI infrastructure. Every time NVIDIA ships a new GPU generation — from H100 to B200 to whatever comes next — the performance gains depend as much on memory bandwidth as on compute. SK Hynix’s HBM3E is currently the fastest memory in volume production, and its next-generation HBM4 is what will feed the next wave of AI training clusters. Locking in a deep, multi-year integration with the supplier is not just a procurement move; it’s a competitive moat.

A Gigawatt AI Cloud for South Korea

SK Telecom’s deal is the most operationally concrete. The Korean telecoms giant will build what it describes as a “gigawatt-scale AI cloud” — a data center complex drawing a full gigawatt of power — using NVIDIA’s platform, with the first facility scheduled to come online in 2027.

To put that scale in context: a gigawatt is roughly the output of a large nuclear reactor, and it represents an order-of-magnitude step up from the 50-100 megawatt facilities that defined enterprise data centers just a few years ago. The frontier AI labs are already deploying at this scale; the SK Telecom announcement signals that national-level AI infrastructure is catching up.

The facility will be designed for “sovereign and agentic AI capabilities” — a phrase that reflects Seoul’s specific concern. South Korea, like many mid-sized economies, faces a choice: build independent AI infrastructure now, at enormous capital cost, or become dependent on US cloud providers for the compute that will increasingly underpin economic and military competitiveness. SK Telecom’s gigawatt cloud is Hwang Changkyu’s answer to that question.

Naver: From Search to AI Factory

Naver, South Korea’s dominant search and internet company, is upgrading its flagship Sejong data center from 55 megawatts to gigawatt scale using NVIDIA’s full-stack Data Center Expansion (DSX) platform. Naver already operates one of the most sophisticated AI stacks outside the US hyperscalers, including its own large language models that power search, translation, and generation services for Korean-language content. At gigawatt scale, Naver transitions from a large data center operator to an AI factory capable of training frontier models domestically.

Physical AI: LG and Hyundai

Two of Korea’s most iconic industrial brands are betting their futures on physical AI in partnership with NVIDIA.

LG Group’s deal focuses on humanoid robotics — a convergence of NVIDIA’s Isaac robotics platform with LG’s manufacturing expertise in motor technology and mechanical systems. Huang framed the collaboration in typically expansive terms: “We are working with them in motor technology as well as mechanical systems so that we can bring together humanoid robotics and the future of robotics.”

Hyundai Motor Group’s partnership spans autonomous vehicles, robotics, and AI-powered manufacturing. Huang singled out Hyundai’s Saemangeum complex — a massive industrial facility being built on reclaimed land on Korea’s west coast — as what he called an “AI Valley.” The ambition is to build AI into manufacturing processes themselves, not just deploy it in vehicles.

Doosan Group rounds out the six, focusing on robotics and materials science applications of AI.

The Sovereign AI Calculus

What unites these deals is less about NVIDIA’s revenue — though six agreements with Korea’s industrial champions will add materially — and more about geopolitics. South Korea sits at an uncomfortable intersection. It is a close US military ally, deeply integrated into US technology supply chains, and simultaneously one of the world’s most important semiconductor manufacturers, with companies (SK Hynix, Samsung) whose customers include both the US and Chinese AI ecosystems.

As the US continues tightening semiconductor export controls targeting Chinese buyers, South Korea faces increasing pressure to choose sides. Building out domestic sovereign AI infrastructure — compute, clouds, models — is partly a hedge against that pressure. If South Korea has its own frontier AI capability, it is less dependent on any single external power for the systems that will increasingly run its economy.

Huang’s visit served that narrative perfectly. He arrived not as a vendor making a sales call, but as a strategic partner lending NVIDIA’s validation to South Korea’s ambition. The message to Seoul: you are not a client. You are a node in the infrastructure that makes AI possible.

Supply Chain Synchronization

Huang left Seoul with one explicit warning. He stressed the need to “synchronize the AI supply chain in the second half of 2026” — code for the growing complexity of coordinating memory production, GPU manufacturing, networking hardware, and facility construction across a global supply chain where any bottleneck cascades. With SK Hynix supplying HBM, TSMC manufacturing chips, South Korean conglomerates building data centers, and NVIDIA providing the connective tissue, the second half of 2026 will test whether that coordination holds at scale.

For South Korea, the Jensen Huang visit is already a success measured in partnerships signed. Whether those partnerships deliver on their ambitions will be measured in gigawatts operational, models trained, and robotics deployed — starting in 2027.