A planned South Korea AI factory is carrying a much larger commercial number than its public build sheet.

Tom's Hardware put Nvidia and SK Group's letters of intent at more than $500 billion, while the disclosed infrastructure anchor is SK Telecom's planned 2-gigawatt AI data centre.

The deal combines three layers: a long-term memory supply agreement with SK hynix, a South Korea facility based on Nvidia hardware and future AI infrastructure expansions.

SK Group's announcement gives 2027 as the first service target and names Nvidia's DSX AI factory platform, Vera Rubin accelerated computing systems and SK hynix HBM4 memory for the installation.

SK Telecom Plans A 2-Gigawatt AI Factory

SK Group's announcement puts the planned South Korea facility at 2 gigawatts and identifies SK Telecom as the operator.

SK Telecom is positioning the buildout for sovereign AI, enterprise AI, physical AI and agentic AI workloads in South Korea and across Asia-Pacific.

The deployment path is still broad rather than contractual.

The public details identify Nvidia's DSX platform, Vera Rubin systems and SK hynix HBM4 memory, but they do not name site location, power provider, construction budget, customer contracts or rack count.

Nvidia's DSX definition combines accelerated computing hardware, networking, software and partner technologies into a data-centre-scale platform.

The planned facility therefore depends on a wider supply chain covering GPUs, networking equipment, systems, memory, cooling, storage and power infrastructure.

SK Hynix Memory Supply Anchors The AI Infrastructure Package

The memory component is central because SK hynix supplies high-bandwidth memory used with advanced AI accelerators.

The new letters of intent follow a June multi-year memory co-development and supply agreement between Nvidia and SK hynix.

The commercial value is not presented as one construction contract.

Anton Shilov interpreted the more than $500 billion figure as aggregate commercial activity over several years, bundling AI infrastructure construction with the long-term memory supply agreement.

Execution detail still determines how the value is allocated.

SK Telecom's purchases of Nvidia GPUs, networking equipment and systems, SK hynix memory supply to Nvidia, partner revenue from DSX AI factory construction and later infrastructure expansion may all contribute to the total, but the public record does not split the value by counterparty or year.

Rack Counts Remain Outside The Public Record

The facility's power envelope points to large infrastructure scale, yet the exact hardware count remains unavailable.

Shilov gave a rough scenario in which a 2-gigawatt AI data centre would use thousands of NVL72 VR200 racks and hundreds; the source also described of thousands of Vera CPUs and Rubin AI GPUs, while warning that the announcement was too vague for an accurate estimate.

Shilov gave NVL72 VR200 rack power ratings of 166 kW to 240 kW and added that DSX deployments can sit in facilities with different power usage effectiveness.

Without the selected rack configuration and facility PUE, the 2-gigawatt figure cannot be translated into a reliable accelerator count.

For Nvidia, SK Telecom and SK hynix, the disclosed plan connects memory supply, AI-factory hardware and a national-scale data-centre project.

The project still lacks the site and configuration evidence needed to link the stated value to named customers or yearly spending.