Samsung Electronics (005930 KS) and SK Hynix (000660 KS) are to mass produce and ship 8-layer HBM4 next February, supplying to Nvidia (NVDA)

Samsung Electronics (005930 KS) and SK Hynix (000660 KS) are to mass produce and ship 8-layer HBM4 next February, supplying to Nvidia (NVDA)

Supply agreements of this kind between Korean memory makers and Nvidia have become the recurring cadence of the AI hardware cycle: each new HBM generation follows the same sequence of qualification, mass production and volume shipment, with the qualification phase historically the point at which timelines slip and market share between the two Korean suppliers and the US rival shifts. The distinction worth drawing is between mass production as scheduled, which locks in near-term revenue visibility for both names, and yield or qualification setbacks, which in prior generations have tended to transfer orders toward the competitor already qualified and to widen the valuation gap between the leaders and the laggard. Co-leadership by Samsung and SK Hynix in the next-generation stack contrasts with earlier rounds in which one supplier held a clear lead, a pattern that in past cycles has compressed the premium the leader commanded and, on the Nvidia side, eased the supply constraint that has governed accelerator shipment schedules. The follow-ons that matter are any read-through from Nvidia on volume commitments and pricing, qualification updates at the next earnings releases, and the cadence of 12-layer variants, since layer count has tended to be where yields and margins diverge. As a production milestone rather than a demand datapoint, the read is about the supply side of the HBM constraint.