Nvidia and SK Group Seal $500 Billion AI Deal — 2GW Factory and HBM4 Memory Supply Locked

Nvidia and South Korea's SK Group announced a $500 billion-plus strategic partnership on July 24 covering a 2-gigawatt AI factory in South Korea and long-term HBM4 memory co-development with SK Hynix, locking in Nvidia's memory supply for the Vera Rubin GPU architecture and positioning South Korea as a sovereign AI compute hub. The deal is the largest single AI infrastructure commitment either company has made.

Nvidia and SK Group Seal $500 Billion AI Deal — 2GW Factory and HBM4 Memory Supply Locked Nvidia and South Korea's SK Group announced a $500 billion-plus strategic partnership on July 24 covering a 2-gigawatt AI factory in South Korea and long-term HBM4 memory co-development with SK Hynix. The deal locks in Nvidia's memory supply for the Vera Rubin GPU architecture and positions South Korea as a sovereign AI compute hub. It is the largest single AI infrastructure commitment either company has made. The Partnership Covers a 2-Gigawatt AI Factory in South Korea and Long-Term HBM Memory Supply — at a Scale No Prior Deal Has Matched Nvidia /glossary/nvidia and South Korea's SK Group announced a $500 billion-plus strategic partnership on July 24 at the AI Summit in San Francisco, spanning a 2-gigawatt AI factory in South Korea and a long-term high-bandwidth memory co-development pact with SK Hynix. The deal locks in Nvidia's HBM4 supply for the Vera Rubin architecture and positions South Korea as the center of gravity for AI infrastructure outside the United States. The partnership is the largest single AI infrastructure commitment either company has ever made. It addresses two existential constraints simultaneously: Nvidia's need for guaranteed HBM supply at a time of global memory shortage, and South Korea's ambition to become a sovereign AI compute /glossary/compute hub rather than a component supplier for American data centers. The Two Pillars of the Deal The agreement has two main components. First, a long-term HBM memory supply and co-development partnership with SK Hynix that secures next-generation HBM4 capacity for Nvidia's Vera Rubin GPU architecture. The global HBM market is supply-constrained — every major GPU manufacturer is competing for the same limited capacity. By locking in SK Hynix production lines directly, Nvidia eliminates the risk of memory bottlenecks delaying future GPU generations. Second, a 2-gigawatt AI factory in South Korea. Two gigawatts is roughly the power consumption of a small city. For context: Colossus 1, the SpaceXAI facility in Tennessee that Anthropic /glossary/anthropic rents for $1.25 billion per month, operates at 300 megawatts. The SK Group facility would be nearly seven times that scale. It is designed as a dedicated AI compute facility — not a general-purpose data center with GPU racks added, but a facility purpose-built for AI training /glossary/training and inference /glossary/inference workloads at unprecedented scale. Why South Korea, Why Now South Korea has been the world's most important memory chip supplier for decades. SK Hynix and Samsung control the majority of global HBM production. But South Korea has historically supplied components — HBM chips — to American companies that build the final AI systems. The $500 billion deal changes that by moving South Korea from component supplier to full-stack AI infrastructure provider. The timing is not coincidental. China is pressuring domestic AI firms to abandon Nvidia silicon. The US continues tightening chip export controls. Japan and Taiwan are building out their own AI infrastructure. Every major economy is attempting to secure domestic or allied compute capacity. South Korea's bet: if you can't control the GPU design, control the memory that makes the GPUs work — and build the factory that houses both. The Memory Supply Chain Reconfiguration HBM4 is the critical component for Vera Rubin, Nvidia's next major GPU architecture. HBM4 provides significantly higher bandwidth than HBM3e, the current standard for H100 and H200 GPUs. Every frontier GPU generation depends on HBM advances. Without guaranteed HBM4 supply, Vera Rubin's timeline slips. With the SK Hynix partnership, Nvidia has locked in priority access to the most advanced memory production lines in the world. The deal effectively pre-sells SK Hynix's HBM4 production capacity to Nvidia. Other GPU manufacturers — AMD, Intel, Chinese domestic producers — will need to secure supply from remaining capacity or compete for Samsung's production. The HBM market was already tight. This deal tightens it further for everyone not named Nvidia. The Geopolitical Dimension A $500 billion AI factory in South Korea also serves a geopolitical purpose. The US wants allied countries to host advanced AI infrastructure rather than see it built in China or under Chinese influence. South Korea is a treaty ally with a mature semiconductor ecosystem. Building the world's largest AI factory there is both commercially rational and strategically aligned with US interests. But it also means a significant portion of global AI training capacity will sit within range of North Korean artillery. The physical security implications of concentrating AI infrastructure in South Korea are not hypothetical — they are factored into every defense planning document covering the Korean Peninsula. The SK Group deal is the largest AI infrastructure commitment of 2026. It secures Nvidia's memory supply. It positions South Korea as a sovereign AI compute hub. And it reminds everyone that the AI race is not just about models — it is about the physical infrastructure that makes models possible. Memory, power, concrete, and steel. Whoever controls those controls the future. Get AI news in your inbox Daily digest of what matters in AI.