The big risk is not whether AI needs memory; it is whether SK hynix can keep its lead valuable as supply scales. If
HBM4 ramps cleanly, Yongin and Cheongju land on time, and
LTAs keep mix elevated, the thesis works. If capacity slips, buyers dual-source faster, or industry supply catches up into a softer AI spending backdrop, SK hynix can revert toward classic memory-cycle economics faster than bulls expect.