Samsung adopts TSMC 3nm for Nvidia NVHBM

Industry reporting on October 18, 2026 indicates that Nvidia (NVDA) has moved to standardise key manufacturing parameters for its next-generation custom high-bandwidth memory, NVHBM. Samsung Electronics, SK hynix (SKHY) and Micron (MU) have been asked to produce NVHBM base dies using Taiwan Semiconductor Manufacturing Company’s 3-nanometre logic process. At the same time, Nvidia requested that the memory core dies for NVHBM be aligned on a sixth-generation DRAM process, known as 1c, across these suppliers.

This approach is intended to bring a common process baseline to the base and core dies used in NVHBM, regardless of supplier. By concentrating base-die production on TSMC’s most advanced logic node and unifying the DRAM node, Nvidia is setting consistent manufacturing conditions for its custom HBM generation. The October 18 reporting frames this as a coordinated effort involving the three main high-bandwidth memory providers.

The same reporting states that Samsung Electronics has agreed to Nvidia’s NVHBM manufacturing requirements after an internal review. As a result, Samsung will rely on TSMC’s 3nm process to fabricate NVHBM base dies destined for Nvidia. This positions Samsung as both a memory supplier and a customer of TSMC’s advanced logic technology for this specific product line.

Despite this shift for NVHBM base dies, Samsung plans to maintain in-house production for standard HBM products. The company will also continue to manufacture custom HBM for other customers using its own facilities. This indicates a dual-track strategy in which Samsung participates in Nvidia’s standardised NVHBM ecosystem while preserving internal control over other HBM offerings.

Reporting further notes that Samsung is evaluating whether its own foundry operations could serve as a fallback for NVHBM base-die production in the future. This contingency planning is focused on the possibility that TSMC’s 3nm capacity may not be sufficient to meet demand. Any such backup role would supplement, rather than replace, the use of TSMC’s 3nm node for NVHBM.

TSMC is identified in the reports as the central manufacturing base for NVHBM base dies, reflecting Nvidia’s requirement that all three memory suppliers use its 3nm technology. The articles highlight that TSMC is expanding its 3nm capacity through additional production lines. These include further build-out at Fab 18 in Tainan and planned 3nm capacity additions at facilities in Arizona and Kumamoto.

These expansion plans are presented as an important backdrop to Nvidia’s NVHBM standardisation, as they influence the availability of advanced-node capacity needed for base-die fabrication. The reported combination of TSMC’s scaling efforts and Samsung’s review of its own foundry as a contingency underscores industry attention to securing enough 3nm output. Together, these developments frame how Nvidia, Samsung, SK hynix (SKHY), Micron (MU) and TSMC are positioning themselves around the next generation of custom high-bandwidth memory.