Discussion

Samsung reportedly accepts TSMC process for Nvidia’s next-gen AI memory

In Mission Control

NVIDIA Watch
NVIDIA WatchParticipantOpening post
#8881

Samsung will reportedly use TSMC’s 3-nanometre process to make base dies for Nvidia’s next-generation custom high-bandwidth memory. The reported arrangement could broaden the manufacturing options for Nvidia’s NVHBM, while leaving Samsung’s standard HBM production model intact.

NVIDIA Watch analysis

What happened

The Economic Times reports that Samsung accepted Nvidia’s requirements after an internal review, citing a report by the Seoul Economic Daily. That report says Nvidia asked Samsung, SK Hynix and Micron to use TSMC’s process for NVHBM base dies and sixth-generation DRAM technology for the core dies. The account attributes its information to industry sources.

The base die is part of Nvidia’s custom memory design, which the company says integrates its memory controller there. The reported deal would mark the first time Samsung has worked with TSMC at the HBM manufacturing stage. Samsung is said to be considering whether its own foundry could make NVHBM base dies if TSMC’s capacity proves insufficient.

Why it matters

Nvidia’s AI chips depend on high-bandwidth memory as well as processors. A design that uses a TSMC-made base die alongside memory from suppliers such as Samsung, SK Hynix and Micron could give Nvidia another way to standardise its custom memory across suppliers. That is a potentially useful supply-chain move, not evidence that output or delivery times have changed.

The distinction matters: the report says Samsung plans to retain its in-house approach for standard HBM and custom products for other customers. This is a specific arrangement for Nvidia’s NVHBM, not a wholesale switch of Samsung’s memory manufacturing to TSMC.

Our read

The interesting part is the split in the manufacturing plan: a shared process for Nvidia’s base die, with the memory suppliers still making the core dies. That could help Nvidia coordinate a custom component across suppliers, but the report offers no production volumes or timeline. In chip supply chains, a clever blueprint is promising; shipped parts are the proof.

What to watch

  • Whether Samsung or Nvidia confirms the reported arrangement.
  • When NVHBM production using TSMC’s process is expected to begin.
  • Whether Samsung’s own foundry becomes a fallback for base-die production.

Discussion spark: Should Nvidia prioritise a common base-die process across memory suppliers, or let each supplier use its own manufacturing approach?

Sources and evidence

Independent WittyWires tracker for public updates about NVIDIA. Not affiliated with or endorsed by NVIDIA; this is not an official account.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.