Samsung to Tap TSMC's 3nm Process for Producing NVIDIA HBM Base Dies
nashnova research
Samsung Electronics will, for the first time, tap TSMC's 3nm node to build base dies for Nvidia's custom HBM chips — a move that signals Nvidia's control over the memory supply chain now extends to dictating which foundry its suppliers must use.
What exactly did Nvidia demand?
Nvidia told all three memory makers — Samsung, SK Hynix, and Micron — to use TSMC's process for producing NVHBM base dies.
Core dies — the layers that actually store data — must use sixth-generation DRAM technology.
This means → Nvidia is no longer just setting design specs; it is choosing who manufactures the silicon and on which node.
What is NVHBM, and why embed a memory controller in the base die?
NVHBM is Nvidia's next-generation custom high-bandwidth memory, unveiled in August this year. Its key innovation: an Nvidia-designed memory controller integrated into the base die — the bottom layer of the HBM stack that talks to the GPU.
In plain terms = each memory vendor used to have its own "handshake" protocol with the GPU. Now Nvidia wrote a single unified protocol and baked it into the chip's foundation.
This reflects Nvidia's goal: regardless of whose memory dies sit on top, the interface to the GPU must be identical.
Why did Samsung accept — and what did it keep?
Seoul Economic Daily, citing industry sources, reports that Samsung accepted Nvidia's terms after an internal assessment.
Samsung drew a boundary: standard HBM products and custom HBM for other clients stay in Samsung's own fabs.
This means → Samsung's concession is scoped — it hands over foundry choice only on the NVHBM product line; everything else remains in-house.
What happens if TSMC can't keep up?
Samsung is evaluating a fallback: whether its own fabs can take over base-die production if NVHBM demand exceeds TSMC's 3nm capacity ceiling.
In plain terms = Samsung's role could shift from "forced outsourcer" to "overflow backup factory."
This reflects a critical variable for the entire supply chain — where TSMC's 3nm capacity tops out will determine how much share Samsung's fabs can reclaim.
What does this mean for the market?
Nvidia's grip on the HBM supply chain has deepened another layer: from chip design to foundry designation.
Samsung's own fabs face a real-world test of advanced-node competitiveness: even Samsung's memory division is outsourcing to TSMC.
This means → two nodes to watch going forward: how TSMC allocates 3nm capacity, and whether Samsung's fabs can qualify as an NVHBM backup source.
市场有风险，内容仅供研究参考，不构成投资建议。