Nvidia Locks In HBM4 Supply With SK Hynix, Kicks Off 1-GW ‘AI Factory’ Build in Korea

Chip shortages re-ordering industry alliances is not new: in 1982 NEC inked decade-long take-or-pay DRAM deals with IBM when 64-Kb chips were scarce, and in 2006 Intel and Micron created IM Flash to secure NAND for mobile. Today’s HBM bottleneck is similar but at AI-scale; only SK Hynix, Samsung and Micron can stack >12-layer HBM4 at acceptable yields, giving Korea leverage just as memory shifts from commodity to strategic asset. The pact signals a move from the Fabless + Foundry era toward co-designed “AI factories” where compute, bandwidth and power are architected as one system—echoing Edison’s 1890s power stations that underpinned electrification. If AI does become as foundational as electricity or the internet, locking in memory supply in 2026 could matter more in 2126 than today’s share-price gyrations; conversely, the concentration of know-how in a single geography introduces political-economic fragility reminiscent of 1973’s oil-shock dependence on OPEC.