Nvidia, SK Group ink $500B AI infra deal
SK Telecom will build a 2GW Vera Rubin AI factory using SK Hynix HBM4 memory, with the first facility online in 2027, under a new letter of intent.
Nvidia and South Korea’s SK Group signed letters of intent on July 25 for a partnership the two sides value at more than $500 billion, covering AI data centers, memory supply, and next-generation computing infrastructure, per Nvidia’s own newsroom release and corroborating CNBC reporting. SK Telecom will build a 2-gigawatt AI factory in Korea on Nvidia’s DSX full-stack architecture, running Nvidia’s next-generation Vera Rubin accelerated computing platform, with the first facility planned to come online in 2027. SK Hynix separately locks in as a long-term supplier of the high-bandwidth memory, specifically HBM4, that platform needs.
Nvidia frames the deal as addressing AI demand across the Asia-Pacific region; a 2GW facility would rank among the largest single AI compute installations anywhere once it ships.
What it means for operators
This lands two days after Nvidia’s own Vera Rubin cost-per-token pitch, and it’s a companion move to its reported $250 billion OpenAI Ohio backstop: Nvidia is locking in both compute demand and the memory supply chain behind it in the same week. Naming SK Hynix specifically, rather than sourcing HBM4 on the open market, tells you Nvidia sees memory availability as more of a bottleneck than fab capacity for GPUs. If your organization is planning multi-year compute contracts, watch HBM lead times and pricing as a leading indicator of Vera Rubin-era availability, not just GPU allocation, and expect South Korea’s broader AI chip investment push to keep accelerating as a direct result.