NVIDIA is deepening its memory supply chain with Samsung, and the headline is simple: Samsung has been certified to supply HBM4 chips for NVIDIA’s upcoming Vera Rubin platform. That gives NVIDIA another heavyweight partner as it prepares its next wave of AI hardware, while Samsung gets a firmer seat at the table in a market where memory vendors live and die by qualification.
The move also fits a broader pattern in high-end AI silicon: no single company wants to bet its future on one supplier, especially when demand is rising fast and qualification cycles are brutal. Samsung is already working with NVIDIA on other chip projects, including self-driving chips and Groq AI accelerators, which suggests the relationship is becoming more strategic than transactional.
Samsung HBM4 supply for NVIDIA Vera Rubin
Jun Young-hyun, Samsung Electronics’ co-CEO and head of its chip division, discussed the next generation of chips with Jensen Huang in Seoul. The two also talked about long-term cooperation around high-bandwidth memory, including HBM4 and HBM5, which is exactly where the real money is in AI infrastructure right now. Whoever controls supply of these stacked memory parts controls a lot of the pace of the AI buildout.
HBM has become one of the industry’s most contested bottlenecks, with NVIDIA, SK hynix, and Micron all pushing hard to keep pace with increasingly hungry accelerators. Samsung’s certification matters because it turns a hopeful vendor into an approved one, and that is the difference between being part of the roadmap and watching from the lobby.
Samsung’s other chip work with NVIDIA
Beyond memory, the companies are also collaborating on self-driving chips and Groq AI accelerators. NVIDIA says the LP30 Groq AI inference processor will be produced by Samsung and shipped later in the year. That is a reminder that the partnership is not limited to one socket on one board; it stretches across multiple chip categories where manufacturing scale matters as much as design bragging rights.
The broader takeaway is that NVIDIA is widening its manufacturing and supply options just as AI systems become more dependent on specialized memory and inference hardware. If Samsung can deliver on HBM4 for Vera Rubin and keep winning more advanced packaging work, it could become far more central to NVIDIA’s stack than a one-off supplier ever would.