Jensen Huang wins Van Fleet Award as Nvidia, South Korea trio reunites

The Van Fleet Award is diplomatic theater. The structural signal is the tripartite shift from HBM supply contracts to architecture co-design among Nvidia, Samsung, and SK. Technically, Samsung's 2nm GAA node coupling with Nvidia's Rubin roadmap implies a custom AI accelerator path that bypasses TSMC by 2027. SK Hynix's HBM4 16-hi stack interconnect specs are already aligned with next-gen NVLink — this is standard co-definition, not procurement. Compliance exposure deepens: Samsung's Xi'an NAND and SK Hynix's Wuxi DRAM sit in the US entity-list grey zone. Escalating US-China decoupling forces Korean fabs into binary compliance, inflating capex 15-20%. Competitively, AMD's MI350 and Intel's Falcon Shores lose the Korean memory-plus-foundry bundle. Hyperscaler custom silicon (TPU v6, Trainium 2) becomes the sole alternative to the Nvidia-Korea stack. 12-24 month tail: expect a formalized Korean AI Manufacturing Consortium — HBM4 standardization, co-packaged optics, and potentially a joint AI chip P&L entity. TSMC's roughly 70% AI accelerator foundry share faces its first structural challenge.