Goldman's Samsung model is a pure price story — 373.5 trillion won operating profit in 2026, up 8x from 2025. That's roughly $278 billion at current exchange rates. Most of it is conventional DRAM, not HBM.
The breakdown: Conventional DRAM adds 245.9 trillion won. NAND adds 87.1 trillion. HBM only contributes 12.2 trillion in 2026, then 58.6 trillion in 2027.
DRAM operating profit is forecast up 881% to 287.5 trillion won in 2026. Blended selling price up 285% to $1.7 per Gb equivalent. Margin expands from 40% to 81%.
NAND goes from 2.7 trillion won to 89.8 trillion. Price up 279% to $0.5 per 16Gb equivalent. Margin from 9% to 66%, even as Samsung cuts wafers to favor high-density enterprise storage.
HBM revenue is projected at $81 billion in 2027, up 321%. Blended ASP above $4.0 per Gb. HBM's share of DRAM sales rises from 8% to 20%.
Memory is expected to more than offset pressure in mobile and networks. Risks include sudden oversupply, a macro slowdown that loosens supply, mobile margin compression, and execution on HBM4 qualification and long-term contracts.
2026 is a price number. 2027 is where HBM has to show up.
If DRAM margin goes from 40% to 81% on a 285% price jump, how much of that $278 billion survives a 20% ASP cut? That's the real question. This is a cycle trade, not a structural one — until HBM proves otherwise.
The breakdown: Conventional DRAM adds 245.9 trillion won. NAND adds 87.1 trillion. HBM only contributes 12.2 trillion in 2026, then 58.6 trillion in 2027.
DRAM operating profit is forecast up 881% to 287.5 trillion won in 2026. Blended selling price up 285% to $1.7 per Gb equivalent. Margin expands from 40% to 81%.
NAND goes from 2.7 trillion won to 89.8 trillion. Price up 279% to $0.5 per 16Gb equivalent. Margin from 9% to 66%, even as Samsung cuts wafers to favor high-density enterprise storage.
HBM revenue is projected at $81 billion in 2027, up 321%. Blended ASP above $4.0 per Gb. HBM's share of DRAM sales rises from 8% to 20%.
Memory is expected to more than offset pressure in mobile and networks. Risks include sudden oversupply, a macro slowdown that loosens supply, mobile margin compression, and execution on HBM4 qualification and long-term contracts.
2026 is a price number. 2027 is where HBM has to show up.
If DRAM margin goes from 40% to 81% on a 285% price jump, how much of that $278 billion survives a 20% ASP cut? That's the real question. This is a cycle trade, not a structural one — until HBM proves otherwise.