Nvidia and SK Group jointly disclosed a $500 billion-plus initiative spanning AI data center construction and a memory supply partnership, according to a Reuters report.

The headline figure ranks the programme among the largest private-sector AI infrastructure commitments disclosed to date, eclipsing prior multi-year capacity disclosures from competing hyperscalers and chipmakers.

What does the $500 billion-plus cover?

The announced envelope comprises two interlocked components: an AI data center build-out and a memory supply partnership. Together they address the principal binding constraint on AI accelerator rollouts — high-bandwidth memory (HBM) supply — alongside the physical infrastructure required to host dense GPU clusters.

SK Group's role is anchored in SK Hynix, the group's listed memory subsidiary. SK Hynix manufactures DRAM and NAND and supplies HBM3 and HBM3E used in Nvidia's H100 and B-series accelerators through 2025.

Why does the memory link matter?

HBM stacks vertically through TSV-processed DRAM dies on a base logic die, delivering bandwidth in the terabyte-per-second range per stack. The memory content of a current-generation AI accelerator accounts for a meaningful share of the bill of materials, making HBM output a gating factor on accelerator throughput.

SK Hynix qualified first on HBM3 for Nvidia's H100 platform in 2022 and has held the dominant share of HBM3E shipments through 2025, ahead of Samsung and Micron.

What does the data center component add?

The infrastructure leg references new AI data center campuses, with power and cooling architectures tuned to dense accelerator racks. Nvidia's reference designs for current-generation platforms call for liquid cooling at 40-70 kW per rack, escalating to 120+ kW in next-generation configurations.

SK Group operates construction, energy, and industrial subsidiaries. Pairing these with Nvidia's hardware stack enables an integrated campus model spanning land, power, shell, and compute.

How does this reshape the supplier map?

The agreement lifts an existing SK Hynix–Nvidia relationship into a group-level framework that incorporates construction, power, and memory. It positions SK Group alongside Foxconn, Quanta, and Wistron as a system integrator for AI campus builds.

Competition in HBM intensified through 2024-2025, with Micron qualifying HBM3E for Nvidia platforms and Samsung pursuing qualification for next-generation products. Multi-source qualification has shifted pricing power back toward chip designers and hyperscaler buyers.

What remains undisclosed?

The Reuters dispatch does not specify the phasing of capital outlays, the geographic split between Korea and the United States, or the share allocated to data center capex versus memory expansion. Specific site selections, grid connections, and water-use arrangements for liquid-cooled campuses also remain unspecified.

Nvidia and SK Group did not break out the headline figure into component disclosures at the time of announcement.