KB News Desk
Korea’s tech leaders deepen ties with Nvidia across chips, AI factories and cloud infrastructure
The heads of Samsung Electronics, SK Group and Naver are working to meet Nvidia Chief Executive Jensen Huang near Silicon Valley this week, industry sources said Wednesday.
The meeting is likely to take place Friday, with Samsung Chairman Lee Jae-yong, SK Chairman Chey Tae-won and Naver Chairman Lee Hae-jin in attendance. OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei may also join, though none of the companies has confirmed the meeting or its guest list.
It would be the second such gathering in six weeks, with the venue reversed. Huang spent four days in Seoul in early June meeting SK, Naver, LG Group, Hyundai Motor Group and Doosan. Nvidia used the visit to disclose expanded agreements with each. This week’s meeting appears aimed at moving that work forward rather than opening anything new.
The date also puts the chairmen in the United States on the day SK hynix reports second-quarter earnings. Its guidance on high-bandwidth memory pricing and capital spending arrives as memory and chip stocks are down by double digits from their recent highs.
The three companies are not bringing the same agenda.
SK has the most to protect. Its memory co-development deal with Nvidia widened in June to span four platforms covering AI infrastructure, personal AI and physical AI. SK Telecom plans a gigawatt-scale AI cloud service in Korea using Nvidia’s DSX platform. Chey said in June that the group aims to finish its first domestic AI factory in 2027 and build in Japan later in the decade.
Naver is earlier in the same business. The company plans to construct an AI factory that pairs graphics processing units with data center space, power and cooling, then sells the resulting computing capacity as a service. Under the plan Nvidia disclosed in June, Naver will expand its Gak Sejong data center to 55 megawatts and eventually toward gigawatt scale.
Naver has also joined Nvidia’s Nemotron Coalition for open model development as the first Korean member, and plans to launch an AI agent platform at home in the second half.
Capacity that large needs anchor tenants. OpenAI and Anthropic are the sort of customers who could fill it, which is why their possible attendance matters beyond the guest list.
Samsung’s interest runs through foundry work, meaning contract manufacturing of chips designed by other companies. Samsung trails Taiwan’s TSMC by a wide margin in that business, and Nvidia’s ecosystem has recently begun to open to it. Samsung already makes the Groq 3 LPU, a low-latency inference accelerator that pairs with Nvidia’s Vera Rubin systems.
Jun Young-hyun, vice chairman and head of Samsung’s Device Solutions division, which runs the company’s memory and foundry operations, said his meeting with Huang on June 8 covered next-generation foundry work along with HBM4E and HBM5. Lee Jae-yong had skipped the dinner Huang held with the other chairmen three days earlier.
Reports in early July citing The Information said Anthropic is in early talks to use Samsung’s 2-nanometer process for its own AI chips. Neither company has confirmed it.
On memory, both Korean suppliers ship HBM4 for Vera Rubin, which Nvidia said reached full production in late May.
