Samsung Electronics will manufacture base dies for Nvidia’s next-generation high-bandwidth memory using Taiwan Semiconductor Manufacturing Company’s 3-nanometre process, the Seoul Economic Daily reported on Sunday, citing industry sources. If accurate, the move would be the first time Samsung has relied on TSMC’s process technology at the HBM production stage.
Why this matters
The reported agreement touches several strategic areas in the semiconductor sector: supply-chain resilience for advanced memory used in artificial intelligence accelerators, closer technical alignment between memory suppliers and a leading GPU designer, and an unusual cross-company use of foundry process technology.
Nvidia’s NVHBM product family, revealed publicly in August, integrates a custom memory controller into the base die. The design aims to standardise how memory is implemented across multiple suppliers, allowing companies such as Samsung, SK Hynix and Micron to ship compatible modules while using their own DRAM stacks for the core dies.
- NVHBM integrates a memory controller into the base die to standardise implementations.
- Nvidia has reportedly asked suppliers to adopt TSMC’s 3nm process for base dies and sixth-generation DRAM for core dies.
- Samsung will keep producing standard HBM and bespoke memory for other customers under its in-house model.
What the Samsung decision implies
The Seoul Economic Daily said Samsung accepted Nvidia’s technical requirements after an internal review, but will preserve its existing in-house manufacturing approach for standard HBM products and for custom orders from other customers. The report added Samsung is evaluating whether its own foundry could handle NVHBM base-die production if demand outstrips TSMC’s capacity.
That last point highlights a recurring tension in semiconductor supply: advanced process nodes are scarce and foundry capacity limited. Using TSMC’s 3nm process for a base die allows memory makers to meet Nvidia’s specification quickly, while preserving each supplier’s ability to build proprietary DRAM stacks on their preferred process nodes.
| Company | Role in NVHBM (reported) |
|---|---|
| Nvidia | Defines NVHBM spec; integrates custom memory controller into base die |
| Samsung Electronics | To use TSMC 3nm for NVHBM base dies; retain in-house production for other HBM |
| TSMC | Provide 3nm process technology for base-die manufacture |
| SK Hynix, Micron | Also asked by Nvidia to adopt TSMC 3nm for NVHBM base dies |
Industry context and consequences
The shift is notable because foundries typically guard process know-how closely. Samsung and TSMC are among the very few companies able to manufacture at advanced nodes like 3nm, and they are usually competitors for high-margin logic and system-on-chip business. The reported use of TSMC process technology as a base for products made by Samsung suggests pragmatism in the face of rapid demand from hyperscalers and AI hardware vendors.
For customers and governments watching chip supply chains, a few broader consequences follow directly from this arrangement:
- Faster rollout of NVHBM-equipped accelerators because multiple memory suppliers can meet a shared specification.
- Potential pressure on TSMC capacity, which could prompt Samsung to readjust its internal foundry roadmap to provide fall-back capacity.
- Increased standardisation across memory vendors, which can simplify system integration for AI and HPC customers.
The report is based on unnamed industry sources and was relayed by Reuters. Samsung’s longer-term strategy, according to the same report, remains to balance partnerships with independent in-house production so it can serve other clients without depending solely on external foundries.
As the market for AI accelerators and the memory technologies that support them grows, such cross-company technical arrangements are likely to increase. They underline how hardware makers are willing to mix and match process capabilities to meet tight technical specifications and surging demand.
