SK Group and NVIDIA sign $500 billion AI infrastructure and memory partnership

SK Group and NVIDIA announced an expanded strategic partnership on July 24, 2026 valued at more than $500 billion, covering both new AI data center infrastructure in South Korea and a long-term supply agreement for next-generation memory. The announcement came at an AI summit in Seoul hosted by South Korean President Lee Jae Myung, part of a broader roughly $950 billion set of semiconductor agreements unveiled at the same event — including the separate $200 billion Samsung-Broadcom memory and foundry deal announced the same day.

A 2-gigawatt AI factory built on NVIDIA's full stack

SK Telecom, part of SK Group, will build a 2-gigawatt AI factory in South Korea using NVIDIA's DSX platform and Vera Rubin accelerated computing hardware. The facility is projected to begin operations in 2027 and will integrate NVIDIA's accelerated computing, systems, and software into what the companies describe as a full-stack AI factory architecture optimized for cost and energy efficiency. NVIDIA and SK frame the project as supporting sovereign, physical, agentic, and enterprise AI services across the Asia-Pacific region — a deliberately broad set of workloads rather than a single specialized use case.

SK Hynix locks in roughly 70% of NVIDIA's HBM4 allocation

The memory component of the deal is arguably the more consequential piece. NVIDIA and SK Hynix, an SK Group affiliate, signed a long-term partnership to co-develop and supply next-generation High Bandwidth Memory, with SK Hynix expected to hold approximately 70% of NVIDIA's HBM4 allocation for the Vera Rubin platform. HBM sits directly next to AI accelerator dies, feeding them data fast enough to keep pace with compute — and it has been one of the most supply-constrained components in AI hardware for the past two years as every accelerator maker competes for the same limited manufacturing capacity.

Securing 70% of NVIDIA's HBM4 allocation for a single platform generation is a significant commercial win for SK Hynix over rivals Samsung and Micron, both of which also supply HBM to NVIDIA. It effectively locks in SK Hynix as NVIDIA's primary memory partner for the Vera Rubin generation specifically, even as Samsung separately expands its own memory and foundry relationship with Broadcom.

Why South Korea is positioning itself as an AI infrastructure hub

SK Group Chairman Chey Tae-won said that in the AI era, competitiveness depends on the capacity to produce intelligence at scale, framing the deal as part of Korea's push to become a global AI hub rather than purely a components supplier. NVIDIA CEO Jensen Huang pointed to South Korea's existing strengths in networking, data centers, chip manufacturing, and industrial scale as reasons the country is positioned to become what he called a global AI powerhouse.

The scale of the day's combined announcements — Samsung-Broadcom at $200 billion, SK Group-NVIDIA at over $500 billion, plus additional smaller deals including a $10 billion NAVER-Brookfield-NVIDIA AI factory expansion — signals a coordinated push by the South Korean government to position the country as critical infrastructure for the next phase of AI buildout, not just a chip-manufacturing base for other countries' AI companies.

What to watch next

The partnership builds on an existing multi-decade relationship between SK Group and NVIDIA rather than starting from scratch, which should smooth execution risk relative to a first-time partnership at this scale. The 2027 target for the first AI factory coming online gives a concrete near-term milestone to track. As detailed in NVIDIA's official announcement, specific contract volumes and pricing for the HBM4 supply agreement were not disclosed.