Samsung's contract prices for high-bandwidth memory (HBM) are projected to more than double by 2027, The Korea Herald reported.

The Korean newspaper surfaced the pricing outlook, which applies to HBM — the vertically stacked DRAM product that sits adjacent to the compute die inside AI accelerators and supplies terabytes-per-second of bandwidth to tensor and matrix engines.

What does "more than double" mean in dollar terms?

The Korea Herald report did not specify a baseline price, a percentage figure, or a target dollar amount. Without a starting reference, the absolute magnitude of the increase is not quantified.

For context, recent HBM3E contracts have priced 8-high 24 GB stacks in the low-to-mid hundreds of dollars per stack depending on volume tier and qualification status. A doubling would push per-stack pricing above $500, with proportional effects on 12-high and 16-high configurations as they ship.

Why is the 2027 horizon significant?

The 2027 window places the projected increase inside the deployment period for next-generation AI accelerators. Planned platforms from NVIDIA's Vera Rubin generation, AMD's MI400 successors, and custom tensor processing units from hyperscalers all rely on HBM4 and subsequent generations during this period.

Memory cost rises with each generation. An H100-class accelerator carries five HBM3 stacks; successors such as B100 and B200 carry eight. Doubling memory cost adds several hundred dollars per accelerator before integrator markup. Across a hyperscaler fleet of tens of thousands of GPUs, the aggregate cost swing reaches into the billions.

What HBM generations does Samsung produce?

Samsung manufactures HBM2E, HBM3, HBM3E, and is qualifying HBM4. The company's HBM3E qualification with NVIDIA proceeded more slowly than competitors during 2024 and early 2025, affecting allocation in the H200 and B100/B200 supply cycles.

SK hynix has held the leading position in HBM3E shipments, with Micron Technology competing for secondary qualification slots. The three suppliers — Samsung, SK hynix, and Micron — collectively cover the merchant HBM segment.

What could limit the projected doubling?

Several factors could compress the increase:

• Qualification setbacks at Samsung that shift volume back to SK hynix or Micron
• Faster HBM4 ramps at competitor fabs that ease supply tightness
• Hyperscaler demand softening during 2026-2027
• Architecture shifts that reduce HBM stacks per accelerator

Hyperscalers typically negotiate multi-quarter supply contracts with price caps. A doubling would test those contract structures and could accelerate second-source adoption.

What did the Korea Herald report specify?

The Korea Herald cited industry sources for the pricing outlook. The report did not indicate whether the projection applies across all HBM generations or to specific tiers such as HBM3E versus HBM4.

Samsung, SK hynix, and Micron have not publicly commented on 2027 contract pricing.

What data points will clarify the trajectory?

Three indicators will show whether the projection holds:

1. Samsung's HBM3E qualification status with NVIDIA through the second half of 2025
2. HBM4 contract pricing as initial agreements are signed in late 2025
3. Capital expenditure announcements from SK hynix and Micron for HBM4 capacity expansion

The Korea Herald report offers a directional signal. Contract negotiations through 2026 will determine whether buyers see the projected pricing or pay accordingly.