Convequity on X
We see $NVDA seeking more power over its HBM suppliers by making Samsung, SK hynix and Micron ($MU) conform to the same base die design for NVHBM, its custom memory. Seoul Economic Daily reports Nvidia asked all three to make the NVHBM base die at TSMC ($TSM), citing consistent performance across suppliers and less validation work per supplier. Samsung reportedly agreed. The base die is the logic beneath the DRAM stack that connects it to the processor, and it is where each memory maker could differentiate with its own logic design. On NVHBM the base die follows Nvidia's standard and carries Nvidia's own memory controller. Our view: this makes the three suppliers' NVHBM stacks much closer to interchangeable. Nvidia can validate each supplier's memory more easily, shift orders between them and push harder on price. The trade-off is more reliance on TSMC. The memory makers are left competing on DRAM core dies, stacking, yield and capacity, which weakens their pricing power with Nvidia on NVHBM. What would change our view: SK hynix or Micron keeping their own logic in the NVHBM base die, or NVHBM staying a small share of Nvidia's memory purchases.
