Samsung turns to TSMC for base dies in Nvidia’s next-generation HBM

Samsung is reportedly preparing to use TSMC’s 3-nanometre process for the base die of Nvidia’s custom NVHBM, marking an unusual cross-company manufacturing deal as AI memory demand intensifies.

Samsung Electronics is set to use TSMC’s 3-nanometre manufacturing process for the base die in Nvidia’s next generation of custom high-bandwidth memory, according to the Seoul Economic Daily. The arrangement would be the first time Samsung has worked with the Taiwanese foundry at the HBM production stage, underlining how Nvidia is pushing memory suppliers towards a more standardised design for its AI systems.

The paper said Nvidia has asked Samsung, SK Hynix and Micron Technology to use TSMC’s process for the base die in its NVHBM modules, while relying on sixth-generation DRAM technology for the core die. NVHBM, which Nvidia unveiled in August, integrates the company’s custom memory controller into the base die, with the aim of making implementation more uniform across multiple suppliers.

Samsung has reportedly accepted Nvidia’s requirements after an internal review, while keeping its own manufacturing model for conventional HBM and for custom products made for other customers. The Seoul Economic Daily also said the company is weighing whether its in-house foundry could build NVHBM base dies if demand outstrips TSMC’s available capacity.

The move reflects the intensity of competition in advanced memory for artificial intelligence accelerators, where performance, power efficiency and supply flexibility have become closely linked. One report cited by South Korean industry sources said the NVHBM design is intended to lift memory bandwidth, reduce power use and free up space on Nvidia’s compute die, while another said the partnership could help address a broader supply bottleneck as demand for high-bandwidth memory rises.