SK Group, Nvidia chiefs talk AI in Silicon Valley: sources

SK Group chairman Chey Tae-won met Nvidia CEO Jensen Huang in Silicon Valley last week, accompanied by SK Biopharmaceuticals vice president Chey Yoon-chung, industry sources said Wednesday.

The two leaders reportedly discussed potential cooperation in AI, including chip supply and biotechnology sectors.

During the two-hour discussion, the focus was believed to be on supply plans for Nvidia’s next-generation high-bandwidth memory, HBM4, used in upcoming AI hardware. The sources also indicated that the leaders explored opportunities to develop South Korea’s AI ecosystem.

Chey Tae-won has been in the US since early February for meetings with major tech firms, according to industry sources.

🔗 Source: Yonhap

🧠 Food for thought

Implications, context, and why it matters.

Beyond chip supply, a massive AI factory partnership is already underway

SK Group chairman Chey Tae-won and NVIDIA CEO Jensen Huang met to talk HBM4 supply, while continuing work already in motion. NVIDIA and SK Group are building an AI factory in Korea with more than 50,000 NVIDIA GPUs, with phase one slated to finish by late 2027 1.
The site is set up to support SK affiliates plus outside groups through GPU-as-a-service, which lets customers rent GPU capacity. SK hynix (an SK Group memory-chip subsidiary) uses NVIDIA tools to speed parts of chip design and to build semiconductor fab digital twins (virtual replicas of factories used to simulate and optimize operations), including NVIDIA Omniverse (a 3D simulation and collaboration platform) and CUDA-X-related tools (NVIDIA’s software libraries for accelerating computing workloads on GPUs) 1.
SK Group sits in the stack beyond parts supply, since it is weaving NVIDIA’s AI platform into day-to-day work. That work runs alongside SK hynix’s role as a major NVIDIA memory technology partner 1.

SK hynix’s leverage signals a power shift in the AI supply chain

AI hardware constraints are tilting bargaining power toward memory makers such as SK hynix 2.
NVIDIA is reportedly set to source about two-thirds of its HBM4 needs for the Vera Rubin platform from SK hynix, with the share understood to be close to 70% 3.
Korea is also pushing large buildouts, with NVIDIA working with Samsung and Hyundai Motor Group on AI-factory-related infrastructure plans that include large-scale GPU deployments 4.
HBM4 needs about three times more wafer space than standard DRAM (dynamic random-access memory, the mainstream memory used in most computers). One 2026 forecast note links that shift to tighter DRAM supply plus higher prices for conventional computer memory 2.