The Artificial Intelligence infrastructure buildout is still running faster than the physical chips required to support it. Across hyperscalers, chip designers, and memory manufacturers, capital spending is surging, yet supply chains remain pinned by one stubborn constraint: high-bandwidth memory. Demand tied to training and inference workloads continues to rise in waves, not increments, pushing the semiconductor industry toward long-duration planning cycles rather than typical boom-bust inventory corrections.
Jensen Huang’s recent Seoul stop underscored the scale of the supply chain challenge. The Nvidia CEO met SK hynix leadership over chimaek, fried chicken and beer, at Kkanbu Chicken, a setting framed as symbolic of high-level Korean tech diplomacy. According to CNBC, Huang said the global memory shortage “is going to persist for several years.” He also teased a possible announcement tomorrow, which was interpreted as a sign that Nvidia and SK hynix could formally expand their cooperation.
Nvidia and SK hynix already sit at the center of the Artificial Intelligence memory ecosystem. SK hynix is the leading supplier of HBM3E and HBM4 used in Nvidia’s Blackwell accelerators and next-generation platforms such as Vera Rubin. A recent UPI report estimated SK hynix could provide roughly 50% to 70% of Nvidia’s HBM4 requirements. That level of concentration reflects both Nvidia’s supplier preference and the limited number of manufacturers capable of producing high-bandwidth memory at leading-edge performance and yield.
An expanded agreement could include multi-year volume commitments, accelerated HBM4 and HBM4E production ramps, yield and process optimization partnerships, and pricing frameworks tied to capacity expansion. Beyond procurement, prior collaboration with SK Group has introduced the concept of Artificial Intelligence factories in Korea, combining manufacturing, simulation, and Artificial Intelligence optimization. SK hynix has separately outlined plans to roughly double memory capacity over a five-year horizon. The broader implication is that Nvidia is seeking multi-year supply visibility while SK hynix locks in structural demand growth for the Artificial Intelligence cycle.
