Memory oversupply? Not happening — even past 2030.
HBM demand is crushing conventional DRAM capacity. The twist: HBM needs far more wafers per bit than DDR5. Already tight.
Now add 3D stacked DRAM integrating directly with logic, scaling around 2029-2030. That could squeeze conventional DRAM supply even harder. Memory companies haven't even locked down which process nodes they'll use. And if the architecture works as expected, 3D stacked DRAM may constrain bit output even more than today's HBM.
Most analysts aren't pricing this in yet. Expect the conversation to shift next year as more companies partner with memory vendors on 3D stacking.
With SK Hynix earnings strong, this structural supply story is worth revisiting. The memory cycle isn't reversing — it's tightening further.
HBM demand is crushing conventional DRAM capacity. The twist: HBM needs far more wafers per bit than DDR5. Already tight.
Now add 3D stacked DRAM integrating directly with logic, scaling around 2029-2030. That could squeeze conventional DRAM supply even harder. Memory companies haven't even locked down which process nodes they'll use. And if the architecture works as expected, 3D stacked DRAM may constrain bit output even more than today's HBM.
Most analysts aren't pricing this in yet. Expect the conversation to shift next year as more companies partner with memory vendors on 3D stacking.
With SK Hynix earnings strong, this structural supply story is worth revisiting. The memory cycle isn't reversing — it's tightening further.