Counterpoint: HBM4 for Rubin Stays at 12 Layers as Memory Demand Holds

Counterpoint Research dismissed rumors that Nvidia's Rubin accelerator would cut HBM4 stacking from 12 to 8 layers, saying most Rubin-based products converge on 12 layers and no 4-layer parts have appeared even in the ASIC market. With Samsung and SK hynix already mass-producing HBM4, high-value memory demand is expected to support both companies' earnings.

• Counterpoint calls memory demand slowdown fears a "misunderstanding"
• Rubin to keep 12-layer HBM4; no 4-layer products seen in ASIC market
• HBM consumes over three times the capacity of commodity DRAM
• Samsung Q3: revenue 195 trillion won, operating profit 107.4 trillion won