Nvidia Chief Executive Jensen Huang will meet Samsung Electronics (005930) Chairman Jay Y. Lee and SK Group (034730) Chairman Chey Tae-won again in New York on the 28th, local time. The gathering comes about two months after the three met in San Francisco in July. With the heads of companies central to the global artificial intelligence chip supply chain convening just after the U.S.-China summit, attention is turning to whether cooperation on next-generation high-bandwidth memory (HBM) and AI infrastructure will expand further.
Lee and Chey plan to attend the Korea Society's annual gala in New York at 6 p.m. on the 28th, according to business sources on the 27th. Huang will attend as this year's recipient of the Korea Society's Van Fleet Award, which honors individuals and organizations that have contributed to the U.S.-Korea relationship.
In explaining Huang's selection, the Korea Society specifically cited his partnerships with Korean companies including Samsung Electronics, SK Group and Hyundai Motor (005380) Group. Huang, who has led the AI and semiconductor industries, has built long-term partnerships with Korean firms that elevated the country into a core partner in global AI infrastructure, the organization said.
HBM is the industry's chief interest in the meeting. As AI accelerator performance rises sharply, HBM — which rapidly feeds large volumes of data to graphics processing units — has become a core component determining the performance of Nvidia's next-generation products. HBM4, the sixth generation, is used in the Vera Rubin AI accelerator that Nvidia put into full production this year.
Samsung Electronics and SK hynix (000660) are moving beyond supplying HBM4 to Nvidia and are building a framework to jointly develop and optimize successor products including HBM4E, the seventh generation, and HBM5, the eighth. AI chips are shifting away from a model in which memory is selected after the GPU design is fixed, toward co-design in which GPUs and HBM are matched from the earliest design stage. How early a memory maker gains access to Nvidia's future GPU roadmap now determines its competitiveness.
Samsung Electronics has been accelerating its cooperation with Nvidia in this area. Jun Young-hyun, vice chairman and head of Samsung Electronics' Device Solutions division, met Huang during the Nvidia chief's visit to Korea in June and discussed long-term cooperation on next-generation HBM including HBM4E and HBM5, according to sources. Samsung began shipping HBM4 in mass production in February and started supplying 12-layer HBM4E samples to customers in May. The company plans to expand into custom HBM, designing memory performance, power and interfaces to match a customer's GPU architecture.
The New York meeting could also touch on foundry services and advanced packaging, another Samsung strength. Jun and Huang are also understood to have discussed cooperation in next-generation foundry services at their June meeting. Samsung is pursuing the AI chip market by offering memory, foundry services and advanced packaging within a single company. That means its relationship with Nvidia could broaden beyond that of an HBM supplier and customer to cover AI chip production as a whole.
Cooperation between Chey's SK Group and Nvidia has already extended beyond memory into AI infrastructure. In July in San Francisco, the two sides announced a plan under which SK Telecom (017670) will build an AI factory of up to 2 gigawatts based on Nvidia's DSX platform and Vera Rubin systems, with phased operation starting in 2027. SK hynix and Nvidia also formed a partnership to jointly develop next-generation AI memory including HBM over the long term and to supply it reliably.
An AI factory is a large-scale data center that links tens of thousands to hundreds of thousands of GPUs to train AI models and deliver services. As competition in the AI industry shifts from the performance of individual chips to vast systems that bundle GPUs, memory, networking, power and cooling, Nvidia is expanding into supplying entire AI factories. SK's strategy is to target that market by combining its telecommunications and data center operations with SK hynix memory.
Samsung Electronics is also building an AI megafactory with Nvidia. The project deploys more than 50,000 Nvidia GPUs to apply AI to chip design, manufacturing processes, equipment operation and quality control, and to expand digital twin technology that replicates an actual chip plant in a virtual space. Some observers expect the scope of cooperation discussed at this meeting to widen beyond HBM to the AI transformation of chip production sites and to physical AI fields such as robotics.
The continuing technology rivalry between the United States and China adds weight to the meeting. With U.S. export controls on advanced AI chips bound for China and China's push to develop its own AI chips, securing stable production and memory supply chains has become more important for Nvidia. For Samsung Electronics and SK hynix, gaining early insight into the medium- and long-term product strategy of Nvidia, which dominates the global AI accelerator market, is a key variable in deciding the scale of HBM investment.
The three also gathered at a global AI event in San Francisco on July 24 to discuss expanding cooperation in AI chips and infrastructure. Since then, Samsung Electronics and SK Group have each firmed up cooperation with Nvidia on next-generation memory and AI infrastructure, so the New York meeting is likely to focus on the progress of projects already agreed and on long-term plans aimed at the period beyond HBM.
"As AI accelerator competition moves through generations, the structure is changing so that GPU makers and memory makers must jointly develop products years in advance," a business source said. "This event is not a formal business negotiation, but because executives leading the global AI supply chain are gathering in one place, it could serve as an occasion to discuss the direction of future cooperation."
