Nvidia Locks 70% of HBM4 Supply to SK Hynix, Deepening Korea Dependency

Published by James Harris on June 8, 2026June 8, 2026

Nvidia confirmed that Samsung, SK Hynix, and Micron have all qualified to supply HBM4 memory for its Vera Rubin platform, with SK Hynix expected to capture 60 to 70 percent of that volume under a new multiyear partnership. Vera Rubin is now in production and scheduled to ship to AWS, Google Cloud, Azure, and Oracle in Q3, with Nvidia citing 10x the agent throughput of its current Grace Blackwell systems.

The timing of Jensen Huang’s Seoul visit is doing a lot of work here. South Korean markets cratered on Monday, the KOSPI briefly triggering a circuit breaker and falling nearly 17 percent from its recent peak, with Samsung and SK Hynix among the hardest-hit names. Huang’s announcements landed as a kind of counter-narrative: even as Korean equity markets signaled stress, Nvidia was publicly deepening its dependency on Korean semiconductor infrastructure. That dependency is not symbolic. HBM4 is not a commodity memory upgrade; it requires co-engineering between chip designer and memory supplier at a level that makes switching costs extremely high. The SK Hynix partnership, which extends into chip design tooling and factory automation, is Nvidia locking in supply-chain architecture before competitors can replicate it.

Three certified HBM4 suppliers sounds like diversification. At 60 to 70 percent share for one vendor, it is not.

The broader implication is that Nvidia’s AI roadmap is now structurally intertwined with Korean industrial policy in a way that creates mutual exposure. If Korean markets continue deteriorating, whether from macro pressure or renewed trade friction, the supply chain Nvidia just publicly celebrated becomes a concentration risk. For AMD and Intel, both working to close the AI accelerator gap, Nvidia’s deep integration with SK Hynix raises the barrier: qualifying a competitive HBM4 supply chain is not just a procurement problem but a multi-year co-development commitment. The partnerships Huang announced with Naver, LG, and Hyundai also suggest Nvidia is building customer lock-in on the demand side in parallel with supply-side entrenchment.

Vera Rubin shipments to hyperscalers are confirmed for Q3 2025, which means HBM4 volume ramp is not a future story. Allocation figures between the three certified suppliers will become a quarterly earnings variable for Samsung and SK Hynix within months.