Nvidia Locks Down SK Hynix's HBM Supply in $500 Billion South Korea AI Deal
Nvidia and SK Group signed letters of intent covering a 2GW Vera Rubin data center and next-generation HBM4 memory, securing critical chip supply for 2027.
Nvidia and South Korea’s SK Group signed letters of intent on July 25 for a partnership valued at more than $500 billion, spanning AI data centers, memory supply, and next-generation compute infrastructure. It’s one of the largest technology partnerships announced to date, and it does double duty: it locks in Nvidia’s memory pipeline while giving SK a guaranteed customer for its most advanced chips.
The centerpiece is a 2-gigawatt AI data center that SK Telecom will build on Nvidia’s Vera Rubin platform, with the first phase targeted to come online in 2027. Separately, Nvidia and SK Hynix are deepening their collaboration on HBM4 — the next generation of high-bandwidth memory that every AI training cluster depends on to keep GPUs fed with data. SK Hynix is already Nvidia’s primary HBM supplier; this agreement extends that lock-in through the Vera Rubin generation.
The deal also reaches beyond SK. As part of the broader South Korea push, Nvidia is expanding work with Naver and Brookfield, growing Naver’s AI data center in Sejong to 200 megawatts and roughly 100,000 GPUs, also built on Vera Rubin. The announcement came out of an AI summit in San Francisco, and it lands the same week Nvidia is separately negotiating a $250 billion financing backstop for OpenAI’s Ohio data center campus — a signal of just how much capital Nvidia is willing to commit to guarantee demand for its next-generation silicon.
For the memory market, this is the story to watch. HBM has been the binding constraint on AI accelerator production for two years running, and whoever controls guaranteed HBM4 supply controls how fast the next generation of GPUs can actually ship. By locking SK Hynix in this deep and this early, Nvidia is pre-empting the kind of memory shortage that squeezed Blackwell and Rubin-generation production. TipRanks and other analysts have flagged the 2027 timeline as the thing to watch — a lot can shift in eighteen months, and letters of intent aren’t binding contracts.
Still, the scale here — $500 billion in commitments, a 2GW data center, and a captive HBM4 supply chain — makes this one of the clearest signals yet that the AI infrastructure buildout hasn’t slowed down, it’s just moved to bigger, longer-dated bets.