Samsung’s reported decision to use TSMC’s 3nm process for the base die in Nvidia’s custom NVHBM would be a notable break from its usual in-house approach—but a limited one. The reports describe a supplier-specific manufacturing choice for Nvidia’s product, not evidence that Samsung is giving up turnkey production across its HBM business. The companies have not publicly confirmed the reported arrangement.
32
35
What Nvidia reportedly wants suppliers to standardize
According to reports citing industry sources, Nvidia asked Samsung, SK hynix and Micron to make NVHBM base dies on TSMC’s 3nm process and use sixth-generation 1c DRAM for the stacked memory dies. Samsung reportedly accepted the request after an internal review.
1
5
The distinction between the two components matters: the base die is the logic component at the bottom of the HBM stack; the core dies are the memory layers above it. A shared process specification gives Nvidia a more consistent platform to work with across suppliers. It does not mean the finished products would be identical, or that every source has independently confirmed the reported requirement.
5
35
Why this is an exception to Samsung’s turnkey model
Samsung’s turnkey approach means it can make both the logic base die and the memory dies itself. Under the reported NVHBM arrangement, TSMC would supply the manufacturing process for the base die while Samsung remains one of the memory suppliers. That would give up in-house production of one component for this product, rather than the whole HBM business.
32
The reporting describes Samsung as continuing to make base dies in-house for its standard HBM line. That points to a selective response to Nvidia’s specifications, rather than a broad switch away from vertical integration.
11
NVHBM puts more emphasis on the base die
Nvidia’s custom design reportedly moves the memory controller from the processor die into the HBM base die, making that component more than a passive foundation for the memory stack.
40 Nvidia has described NVHBM as offering up to 30% more bandwidth and 15% lower HBM power consumption than standard HBM4E; those are stated product claims, not a guarantee of the same gains in every system.
43
The additional logic helps explain why the base die’s manufacturing process is central to the reported deal. But the sources describe a reported request and agreement—not a public technical document establishing the precise performance or production outcomes.
What the reports do—and don’t—establish
The reported arrangement is strategically significant because it suggests Nvidia can shape how competing memory suppliers manufacture a custom product. For Samsung, the tradeoff would be less control over base-die fabrication for NVHBM in exchange for meeting Nvidia’s requested process specification.
1
5
There are important limits to what is known. Reuters reported the account as a Seoul Economic Daily report, and another report says the companies have not confirmed the specific manufacturing arrangements.
32
35 Claims about yield comparisons, a Samsung-process fallback, or the details of TSMC capacity should therefore not be treated as established outcomes on the evidence available here.
The clearest takeaway is that Samsung’s reported move would make its turnkey model flexible, not obsolete: Nvidia’s product requirements could take priority for NVHBM while Samsung continues to use its own manufacturing for other HBM products.