Samsung reportedly to use TSMC’s 3nm process for Nvidia’s next-generation AI memory
The reported plan would mark Samsung’s first use of TSMC technology in HBM production, but the details and mass-production schedule have yet to be officially confirmed by the three companies.
Samsung Electronics is reportedly planning to use Taiwan Semiconductor Manufacturing Co.’s (TSMC) 3nm process to produce base dies for Nvidia’s next-generation custom high-bandwidth memory (HBM). If realized, the plan would mark Samsung’s first use of TSMC process technology in HBM manufacturing and underscore the importance of cross-company cooperation in the AI chip supply chain.
According to a Reuters report published Oct. 18, the South Korean Seoul Economic Daily cited industry sources as saying that Samsung would use TSMC’s 3nm technology to manufacture base dies for Nvidia’s next-generation custom HBM chips. Reuters’ report is based on the Korean media account, and the arrangement remains subject to official confirmation from Samsung, TSMC and Nvidia.
The report said Nvidia has asked memory suppliers including Samsung, SK Hynix and Micron Technology to use a specified base-die process and sixth-generation DRAM technology in their next-generation custom HBM products. Nvidia unveiled the related NVHBM design in August this year, with the aim of integrating a custom memory controller into the base die so products from different memory suppliers can be more easily integrated with Nvidia platforms.
HBM is a key component of AI accelerators, providing high data-transfer capacity for large-model processing and intensive computing. As generative AI, data centers and high-performance computing systems continue to expand rapidly, memory bandwidth, power consumption and packaging integration have become critical factors in AI hardware competition.
For Taiwan’s semiconductor industry, the report again highlights the role of TSMC’s advanced process technology in the global AI supply chain. If Samsung does adopt TSMC technology, it would suggest that even a competitor may rely on TSMC’s manufacturing capabilities for certain product segments, showing that competition and cooperation in the industry can no longer be viewed solely as direct rivalry between companies.
That would not mean Samsung plans to outsource all HBM production to TSMC. The report cited by Reuters said Samsung still plans to maintain in-house production of standard HBM and customized products for other customers, while evaluating whether its own foundry division could produce the relevant base dies if demand exceeds its supply capacity.
Markets will next watch for Samsung, TSMC and Nvidia to confirm further details of the cooperation, including the production schedule for the relevant process, capacity allocation and product adoption. For Taiwan, opportunities for cooperation in advanced processes and the AI chip supply chain remain an important indicator of the semiconductor industry’s competitiveness.