Nvidia Chief Executive Jensen Huang placed particular emphasis on cooperation with Samsung Electronics (005930.KS) while solidifying multiple artificial intelligence (AI) partnerships during his five-day visit to Korea. Discussions are expanding beyond supplying memory such as high-bandwidth memory (HBM) and low-power DRAM (LPDDR) for Nvidia's next-generation AI accelerator "Vera Rubin" to foundry (contract chip manufacturing) cooperation related to custom chips. Industry observers note that while Nvidia maintains cooperative relationships with Korean companies in individual business areas, it is broadening the scope of cooperation with Samsung Electronics across multiple areas of its entire portfolio.
Huang departed Korea by private jet from Gimpo Airport at around 10 a.m. Monday. He showed a wide-ranging itinerary, visiting Korea again seven months after his October trip to attend the Asia-Pacific Economic Cooperation (APEC) forum, his first visit in 15 years, this time for a vacation. The previous day, he carried out a 12-hour push to meet AI partner companies across Seoul and Gyeonggi Province, from Gwanghwamun to Bundang. Even during his family vacation, he pursued business activities in parallel, meeting business partners and friends including SK Group Chairman Tae-won Choi and LG Group Chairman Kwang-mo Koo in succession.
Huang: "Reason for Coming to Korea is Cooperation with Samsung"
Huang specifically singled out Samsung Electronics as Korea's key AI partner. The previous day, he held a private meeting with Jun Young-hyun, Samsung Electronics Vice Chairman and head of the Device Solutions (DS) Division, at the Nvidia-hosted "Korea AI Ecosystem" reception held at the Shilla Hotel in Jung-gu, Seoul.
Vice Chairman Jun met with Huang on behalf of Samsung Electronics Chairman Jay Y. Lee, who was on an overseas business trip, and said, "We have cooperated (with Nvidia) for a long time, and I think we had the best talks (so far)." Huang responded, "This is Korea's Moment, and I am pleased to be able to build future industries with many friends such as Samsung, LG, Hyundai and SK, with whom we have cooperated 100% all along."
Samsung Electronics is expanding its position as a supplier of HBM4 (sixth generation) to be installed in Vera Rubin and LPDDR5X (seventh generation) used in SOCAMM (low-power memory modules). After supplying HBM4 production units for the first time in the industry in February, it also shipped HBM4E (seventh generation) samples for the first time in the industry last month, and last week unveiled an HBM5 (eighth generation) prototype based on an advanced foundry process at "Computex 2026" in Taipei, Taiwan.
Based on this technological superiority, forecasts suggest Samsung Electronics could absorb some of the market share of SK hynix (000660.KS), Nvidia's largest memory supplier. Some industry observers project that Samsung Electronics' HBM4 market share could expand to around 35 percent, exceeding the initial estimate of 25 percent.
"Medium- to Long-Term Joint Development Discussed"... Signaling Expanded Foundry Cooperation
Regarding cooperation with Nvidia, Vice Chairman Jun said, "In the short term, we discussed HBM4 and foundry cooperation related to autonomous driving chips and accelerators," adding, "In the medium to long term, we exchanged views on joint development plans." Industry observers are paying attention to the possibility that the scope of cooperation between the two companies will expand into the "custom chip" field going forward.
First, it is presumed that discussions were held regarding "custom HBM." The bottommost layer (die) of HBM contains a base die, which is a logic (computing) circuit. Up through HBM3E (fifth generation), the base die was produced using a DRAM process, but from HBM4, a 4-nanometer (nm, one billionth of a meter)-class foundry process was applied to implement a complex signal processing system. Nvidia cites customization of the base die as key to resolving the data transfer bottleneck between the accelerator and HBM. In response to these demands, Samsung Electronics' strategy is to introduce a 2-nanometer foundry process to the base die from HBM5 to further enhance its competitiveness in this area.
The company is also broadening the scope of cooperation in the foundry business. Vice Chairman Jun said, "We are also discussing the next generation of cooperation (with Nvidia) for autonomous driving chips needed for 4-nanometer and 8-nanometer and for our own language processing unit (LPU)." Nvidia relies on TSMC for the contract manufacturing of its flagship graphics processing units (GPUs) and central processing units (CPUs), but entrusts Samsung Foundry with its edge (terminal) autonomous driving chip "Drive AGX Thor" and its LPU "Grok." This is interpreted as evidence that Nvidia recognizes Samsung Foundry's technological capabilities and mass-production capacity, beyond mere supply chain diversification.
In particular, regarding observations in some quarters that TSMC would take over production of Grok (currently third generation) from the fourth-generation product, Vice Chairman Jun directly drew a line, expressing confidence that Samsung Foundry would maintain its position as a key production partner in next-generation products as well.
During this visit to Korea, Huang demonstrated his solid cooperative relationship with SK by meeting SK Group Chairman Tae-won Choi three times, and also shared a cooperation plan with Samsung Electronics encompassing memory and foundry, once again highlighting the importance of Korea's AI chip supply chain. Bae Kyung-hoon, Minister of Science and ICT and Deputy Prime Minister, said, "We also discussed plans to hold GTC in Korea under the name 'GTC Korea,'" to the extent that the industry believes cooperation between Nvidia and Korean companies will become even closer following this visit by Huang.
