Nvidia and SK Group have formalized a $500 billion partnership to build a single 2-gigawatt AI factory in South Korea, according to signed letters of intent announced July 24-25. The facility will be powered by SK hynix's HBM4 memory chips.
The 2-gigawatt scale represents a concentration of compute capacity that exceeds what most individual infrastructure operators have assembled in one location. By comparison, major hyperscalers typically deploy data centers in the 100-megawatt to 1-gigawatt range across multiple sites. The signed letters of intent formalize the commitment without yet constituting a binding construction agreement; timelines for groundbreaking and operational readiness remain tied to regulatory approvals and supply chain availability.
The announcement positions the facility as a hub for AI model training and inference across APAC markets. SK hynix has struggled to compete with TSMC and Samsung in advanced chip manufacturing but holds a leading position in HBM memory, which nvidia GPUs require for handling the bandwidth demands of large language models. The partnership locks SK hynix supply into Nvidia's ecosystem while giving Nvidia direct control over a major source of future compute capacity in a critical geopolitical region.
South Korea's government has been competing with Taiwan and the United States for advanced semiconductor manufacturing investment. The deal aligns with Seoul's broader strategy to anchor AI infrastructure domestically rather than rely on imports. Regulatory approval in South Korea typically takes 6-12 months for infrastructure projects of this scale, and the partnership will require grid capacity allocations from Korean utilities that are currently under strain from existing data center buildouts.
Filecoin noted that no single distributed storage facility has concentrated capacity at the 2-gigawatt scale before. Nvidia has not disclosed whether this factory will serve as a captive manufacturing and training hub or function as a commercial AI service offering.
The $500 billion figure reflects the total capital outlay across design, construction, equipment procurement, and operations through the facility's initial operating phase. That scale places it among the largest infrastructure commitments either company has made in a single geography. Nvidia's total capital expenditure across all facilities globally was approximately $60 billion in 2025, making this partnership's stated scope material to both parties' long-term asset allocation.
Major chip and hyperscale players are committing to 2+ gigawatt facilities as AI model training transitions from multi-site deployments to region-locked, single-source infrastructure. If SK Group and Nvidia begin construction within 18 months of signing final agreements, the facility would likely begin partial operations by 2028-2029, positioning it ahead of comparable projects in Taiwan and the United States that are still in planning or early procurement phases.
