NVIDIA and SK Group Unveil $500 Billion Partnership Spanning AI Factories and Next-Generation Memory Chips

NVIDIA's $500 Billion SK Group Deal Secures Every Layer of Its AI Infrastructure

NVIDIA and SK Group, South Korea's second-largest conglomerate 1, signed letters of intent on July 24 for a partnership valued at more than $500 billion spanning AI factory construction and next-generation memory chips 1. Jensen Huang unveiled the package during South Korean President Lee Jae Myung's visit to the United States, alongside expanded agreements with Samsung, Naver, Hyundai, and LG 2.

The deal's components stack into something unusual. SK Telecom will build a 2-gigawatt AI factory on NVIDIA's DSX platform running Vera Rubin accelerated computing, with the first facility planned for 2027 1. Beyond SK Group, Huang disclosed that Samsung Electronics will pursue chip and memory design partnerships with NVIDIA, that NVIDIA will make a large-scale investment in Naver (which Huang described as one of Korea's leading cloud companies), and that the company will expand work with Hyundai Motor Group on autonomous Genesis vehicles and with LG on power generation systems for factories and homes 2.

Memory supply, compute architecture, cloud infrastructure, factory construction, autonomous vehicles, and power generation. All coordinated through Korean partners. All built on NVIDIA platforms.

Any competitor building AI accelerators now faces a rival that has lined up its memory supply, its factory footprint, and its power infrastructure through a single set of partners. Designing a competitive GPU is hard enough. Assembling the entire infrastructure stack around that GPU, from the memory chips it reads to the power systems that feed it, through separate negotiations with multiple vendors, is a different order of problem. NVIDIA just showed what it looks like when one company coordinates all of it at once.

The diplomatic backdrop sharpens the picture. Lee's U.S. visit included meetings with the leaders of NVIDIA, OpenAI, Broadcom, and Anthropic 2. His strategy aims to turn Korea into a trusted AI production base through a multihub semiconductor network spanning the Seoul metropolitan area and regional provinces 2. Huang committed to establishing a joint AI Frontier Lab in Korea, relocating NVIDIA researchers there, and developing a Korean large language model, in his words, "in Korea and for Korea" 2. Korea is embedding itself directly into the U.S. AI stack as an infrastructure and design partner, not just a component supplier.

Then there is the number. The two sides signed letters of intent, not binding contracts 1. The first AI factory is not planned to come online until 2027 1. And the figure sits inside a broader announcement blitz: the Korea JoongAng Daily reports total Korean tech partnerships exceeding $950 billion, including a separate $200 billion Samsung-Broadcom deal and a $750 billion SK Group agreement that includes NVIDIA and other partners 2. Whether $500 billion represents firm capital commitments or a multi-year aspirational framework is not clarified in either source. The SK hynix section of the announcement includes a standard disclaimer that "no part of this press release should form the basis of, or be relied upon in connection with, any contract, commitment, or investment decision" 1.

If the letters of intent become contracts, the competitive mathematics shift for years. A rival pursuing the same breadth would need to negotiate separately with a memory supplier, a cloud partner, and a power-infrastructure company, each on its own timeline, each capable of stalling independently. That assembly process could consume the time it takes NVIDIA to move from Vera Rubin to its next platform generation. Yet the SK hynix section of NVIDIA's own announcement carries a disclaimer that no part of the press release should form the basis of any contract, commitment, or investment decision 1. The announcement projecting $500 billion in partnership value also tells readers, in legal text, not to rely on a word of it. Both messages serve NVIDIA: the headline stakes out the territory, and the disclaimer preserves the option to walk away from any underperforming piece. The rivals now deciding whether to spend years building a parallel stack are responding to an announcement whose own fine print says it might not happen.