SK Group chairman meets Nvidia CEO to discuss HBM supply: sources

SK Group chairman Chey Tae-won met Nvidia CEO Jensen Huang in California earlier this month to discuss HBM supply plans and AI collaborations, according to industry sources.

The meeting reportedly focused on HBM4, the upcoming memory chip expected to be used in Nvidia’s next AI accelerator, Vera Rubin.

SK hynix, a major HBM supplier, said it is producing HBM4 as scheduled but may not fully meet demand, potentially allowing competitors to secure some contracts.

Samsung is also expected to begin HBM4 production later this year. Industry estimates suggest SK hynix will maintain around a 70% share of Nvidia’s HBM4 market in 2026.

The discussions included potential cooperation related to SK Group’s broader AI strategy beyond memory chips.

Chey has reportedly been in the US since early February for meetings with US tech executives.

🔗 Source: Yonhap

🧠 Food for thought

Implications, context, and why it matters.

The HBM4 discussion is a footnote to a much larger AI partnership

HBM4 supply gets attention, yet SK Group and Nvidia have already announced a collaboration to build an “AI Factory” in Korea with more than 50,000 Nvidia GPUs. The first phase is planned for completion by late 2027 1.
The facility will support SK subsidiaries. It will also offer GPU-as-a-service, which means rentable GPU capacity, to outside organizations. SK Telecom is planning an industrial AI cloud in Asia using Nvidia RTX PRO 6000 Blackwell Server Edition GPUs 1.
For memory chip maker SK hynix, the partnership supports faster manufacturing. It plans to use Nvidia’s Omniverse platform, software for building and simulating 3D virtual environments, plus related tools to build digital twins, virtual replicas of real-world facilities and processes, for its semiconductor fabs, chip factories, to improve production ramp-up and operational efficiency 1.

The global HBM race is feeding into Korea’s sovereign AI push

The SK-Nvidia partnership lands as South Korea pushes to deploy more than 260,000 Nvidia GPUs nationwide. The effort aims to support sovereign AI ambitions, meaning AI capabilities and infrastructure controlled domestically rather than relying on foreign providers 2.
That domestic buildout could create an estimated $750 million in new HBM memory revenue opportunities for SK hynix and Samsung Electronics, based on one estimate cited in external commentary 2.
Pressure also comes from supply constraints and a tight memory market. SemiAnalysis, a semiconductor industry research firm, describes a severe multi-year memory shortage environment. Claims that HBM needs “three times” the wafer resources of standard DRAM vary by source and are not established in SemiAnalysis’ public excerpt 3.