Nvidia and SK Group unveil a $500B+ Korea AI buildout — locking in HBM4 and a 2GW data center

Nvidia and South Korea's SK Group unveiled a more-than-$500B AI initiative spanning large-scale data centers and next-generation memory, announced during President Lee Jae-myung's US visit. The core is a long-term partnership with SK hynix to secure next-gen memory for Nvidia and jointly develop HBM for AI training, agents and physical AI; Nvidia will also work with Samsung on chip and memory design. As part of it, SK Telecom plans a 2GW AI data center running Nvidia's Vera Rubin chips and SK hynix HBM4, with the first facility online in 2027. Nvidia, Naver and Brookfield will expand Naver's Korean AI data center, and Nvidia will deepen a Hyundai tie-up to co-develop autonomous Genesis vehicles.

Why it matters

This nails down the two scarcest inputs in AI — HBM and power-dense capacity — in a single national package. Locking HBM4 co-development with SK hynix (and roping in Samsung) secures the memory that gates every accelerator, the same squeeze Micron's record guidance and the Samsung/SK capex surge have been pricing; a dedicated 2GW build on Vera Rubin is a Stargate-scale anchor outside the US. It also extends Nvidia's strategy of financing demand for its own chips into sovereign-scale commitments — the customer, the memory supplier and the data-center operator bound into one initiative. For Korea's memory duo, being designated Nvidia's co-development partners is the demand visibility their fabs are built against.

What to watch

Firm HBM4 volumes and pricing versus the MOU framing, the SK Telecom 2GW site's 2027 timeline, and how much of the $500B headline converts into contracted capex versus aspiration.