Nvidia locks down Korean memory deals as HBM shortage runs through 2028
Nvidia CEO Jensen Huang visited South Korea in early June 2026 and signed a cluster of multi-year deals with SK Hynix, SK Telecom, Naver, and Doosan Group. The agreements span memory supply, AI cloud infrastructure, and the energy systems needed to run it all. With HBM — the high-bandwidth memory that makes AI accelerators fast — sold out well into the decade, locking in a preferred supplier is less a business courtesy and more a structural necessity.
The memory crunch
HBM demand is growing at 80–100% per year. Supply is growing at 50–60%. Only three companies on the planet make it: SK Hynix, Samsung, and Micron. Analyst consensus puts the shortage running through 2028–2029, and per TechTimes, the SK Hynix chairman has said that even doubling capacity won't close the gap before the end of the decade.
Nvidia already spends billions of dollars annually with SK Hynix. Huang was direct about where that's heading: "SK Hynix is Nvidia's largest memory partner. We already buy billions and billions from them every year, and that volume will grow substantially." The new deal, confirmed by RTE, locks in a multi-year supply structure — and pointedly excludes Samsung from the core partnership.
Cloud, power, and the full stack
Memory is only part of the picture. SK Telecom will build a gigawatt-scale AI cloud using Nvidia's DSX platform, with the first AI factory set to come online in South Korea in 2027. Naver, South Korea's dominant internet company, is adding its own data center infrastructure to the arrangement. Together the deals give Nvidia a vertically integrated base: memory from SK Hynix, compute deployment through SK Telecom and Naver, and the energy backbone from Doosan.
Doosan's role is the least glamorous and possibly the most important. The conglomerate supplies energy solutions and produces the copper-clad laminate (CCL) materials used in Nvidia's next-generation Rubin chips. As AI data centers push into gigawatt territory, having a single partner handle both power delivery and board materials removes one more variable from Nvidia's supply chain.
What this means beyond Korea
US hyperscalers — Meta, Google, Amazon, Microsoft — are competing for the same limited HBM allocations. Nvidia's preferential supply arrangement with SK Hynix gives it leverage to guarantee memory access ahead of those customers, not just alongside them. The deals also sideline Samsung in Nvidia's premium accelerator supply chain, at least for now, concentrating the relationship further with SK.
For anyone buying or building on Nvidia hardware over the next two to three years, the Korea deals are a quiet assurance that the chips will actually exist — no small thing when memory supply is the binding constraint on the entire AI industry.