Lee Jae-yong, Chey Tae-won, and Jensen Huang to Meet in New York After Two Months — HBM and AI Infrastructure Cooperation on Agenda - finance.biggo.com

This New York meeting is NVIDIA recalibrating its dual Korean HBM suppliers at the HBM4 inflection point. Samsung trails SK Hynix on HBM3E, but Vera Rubin shifts competition from production speed to thermal-electrical co-optimization in 16-hi stacks — precisely where Samsung's 2nm GAA and HBM4E base-die integration converge. The June HBM5 pre-discussion signals Samsung bets on next-gen interface definition rights, not current share. On compliance, Huang's White House dinner immediately before this trip is no coincidence. Washington threads the needle between sustaining Korean capacity and restricting HBM outflows to China. Samsung's Xi'an and Suzhou lines face tighter export scrutiny, adding 15-20% per-unit compliance cost. SK Hynix's domestic concentration offers short-term insulation, but SK Telecom's 2GW AI factory power procurement becomes a hidden fiscal burden. The wildcard is Micron. NVIDIA cannot lock HBM4 exclusively to two Korean firms — geopolitical hedging is structural. Yet dual Vera Rubin awards compress Micron's next-gen share below 15%, forcing accelerated HBM4E R&D or a pivot to custom AI memory architectures. Over 18 months, HBM share resets: SK defends HBM3E incumbency, Samsung executes a generational leap via HBM4E. The decisive variable: who first achieves HBM5 16-hi production yield in 2026 — a window aligning precisely with the 18-24 month AI datacenter buildout cycle.