Samsung Electronics and NVIDIA have discussed plans to expand their long-term cooperation, including joint development of next-generation artificial intelligence (AI) memory.
On Monday, Samsung’s Device Solutions (DS) division Vice Chairman Jeon Young-hyun met with NVIDIA Chief Executive Officer (CEO) Jensen Huang at the Shilla Hotel in Seoul. After their meeting, Jeon told reporters that they’ve been collaborating for a long time, and he believes this was the most productive discussion yet.
As NVIDIA aims to establish a global AI factory, it’s likely seeking customized memory solutions like high-bandwidth memory (HBM) rather than general-purpose options. This suggests NVIDIA may have requested Samsung’s collaboration on product development.
Jeon led the discussions with NVIDIA in place of Samsung Electronics Chairman Lee Jae Yong, who was away on a business trip.
The two tech giants focused on short-term plans for HBM4 supply and foundry collaboration.
Jeon explained that it discussed short-term strategies for HBM4 and foundry cooperation. It also explored long-term joint development opportunities. He stressed the importance of supplying sufficient HBM4 and SOCAMM this year, adding that they had in-depth talks about future cooperation, including HBM4E and HBM5 for next year and beyond.
Jeon also highlighted potential expansion in foundry cooperation, noting that Samsung currently produces NVIDIA-related chips using 4nm and 8nm processes. It’s also discussing next-generation collaboration plans, he added.
When asked about CEO Huang’s recent mention of SK Hynix as a key HBM supplier, Jeon confidently stated that it’ll focus on the work and let the results speak for themselves.
Samsung has forged a robust partnership with NVIDIA across various domains. The company currently supplies HBM4 with performance exceeding 11.7 Gbps to NVIDIA and has delivered HBM4E samples. Samsung also provides various memory solutions, including SOCAMM2, optimized for NVIDIA’s next-generation central processing unit (CPU) platform.
For NVIDIA’s Vera Rubin platform, Samsung supplies a range of memory solutions, including HBM4 for the Rubin graphic processing unit (GPU), LPDDR5X-based SOCAMM2 for the Vera CPU, and storage solutions such as PCIe Gen6-based PM1763, NVIDIA CMX-compatible PM1753, and PM9D3a for Boot Drive.
In the foundry sector, Samsung is gearing up to produce next-generation AI chips, including the Grok LPU unveiled at March’s NVIDIA GTC, without delays. Talks are ongoing for 2nm technology collaboration.
The partnership extends to building AI infrastructure centered around AI data centers.
Last October, Samsung and NVIDIA announced plans to establish an industry-leading semiconductor AI factory. During March’s NVIDIA GTC, they showcased a video demonstrating the implementation of an Omniverse-based digital twin at Samsung’s Pyeongtaek Plant 1.
The tech giants are collaborating on an Omniverse-based semiconductor AI infrastructure and plan to expand digital twin applications at U.S. production sites.
