Jace(@JaceX90):Bought some SK hynix. $SKHY My reasoning: all three major memory makers are pouring capex into HBM back-end capacity — but none of them is expanding DRAM front-end wafer capacity anywhere near as fast. HBM eats a huge amount of DRAM wafers. So in theory, even two or three years out, the DRAM supply-demand imbalance gets worse, not better. The more wafers get diverted to HBM, the fewer are left to produce conventional DRAM for everything else. And a single AI server needs a massive amount of HBM. HBM is the key bottleneck for token output speed on inference chips. It's not just NVIDIA — AMD, Google, and every big player building inference silicon is scrambling to secure supply. Full deep-dive here:

2026.07.20 12:45

Bought some SK hynix. $SKHY My reasoning: all three major memory makers are pouring capex into HBM back-end capacity — but none of them is expanding DRAM front-end wafer capacity anywhere near as fast. HBM eats a huge amount of DRAM wafers. So in theory, even two or three years out, the DRAM supply-demand imbalance gets worse, not better. The more wafers get diverted to HBM, the fewer are left to produce conventional DRAM for everything else. And a single AI server needs a massive amount of HBM. HBM is the key bottleneck for token output speed on inference chips. It's not just NVIDIA — AMD, Google, and every big player building inference silicon is scrambling to secure supply. Full deep-dive here: