Nvidia and SK Group have signed letters of intent for a Nvidia SK AI partnership that the companies value at more than $500 billion, tying together memory supply, a planned 2-gigawatt AI data center in South Korea and future AI infrastructure projects.
The deal is broad rather than tidy. It includes the multiyear memory supply and co-development agreement between Nvidia and SK hynix that was disclosed in June, plus SK Telecom’s plan to build an AI data center using Nvidia’s latest data center systems. The first facility is scheduled to begin service in 2027, according to the companies.
SK Telecom plans to base the 2 GW site on Nvidia’s DSX AI factory platform and Vera Rubin accelerated computing systems fitted with SK hynix HBM4 memory. The companies said the infrastructure is intended to support sovereign AI, enterprise AI, physical AI and agentic AI workloads in South Korea and across the Asia-Pacific region.
What is in the Nvidia SK AI partnership?
The disclosed package has several moving parts: Nvidia hardware and networking for SK Telecom’s data centers, SK hynix memory supplied to Nvidia, DSX AI factory construction work involving Nvidia’s partner ecosystem, and possible later expansion beyond the first 2 GW deployment.
That is why the $500 billion figure needs careful reading. Based on what the companies have described, it appears to represent the expected combined value of commercial activity over several years, not one clean purchase order with a public parts list. The announcement folds infrastructure construction and memory supply into the same headline number.
For readers keeping score at home, that means Nvidia is not just selling chips into a data center. SK hynix is also a major supplier into Nvidia’s own accelerator road map, because high-bandwidth memory is a core part of modern AI systems. HBM4, the next generation of that memory, is planned for the Vera Rubin systems cited in the SK Telecom deployment.
What is Nvidia DSX?
Nvidia describes DSX as an AI factory blueprint: accelerated computing systems, networking, software and partner-built facility components arranged as a data center-scale platform. In plainer terms, it is Nvidia’s attempt to sell the whole recipe for running large AI workloads, not just the GPUs.
The companies did not disclose enough technical detail to calculate the final shape of SK Telecom’s first facility. A 2 GW AI data center could require thousands of NVL72 VR200 racks and hundreds of thousands of Vera CPUs and Rubin AI GPUs, but the exact count depends on the rack configuration, the power design and the facility’s power usage effectiveness.
Available public figures for NVL72 VR200 rack power range from 166 kW to 240 kW per rack, depending on configuration. Since SK Telecom has not identified the exact version it will deploy, and has not disclosed the facility’s efficiency metrics, any precise rack or accelerator estimate would be guesswork dressed up as math.
The strategic direction is less ambiguous. Nvidia gets a large Asian customer for its next AI factory platform and a long-term memory relationship with one of the industry’s key HBM suppliers. SK Group gets a path to build domestic and regional AI infrastructure around Nvidia systems, while SK hynix stays close to the accelerator vendor driving much of the HBM demand.
This story draws on original reporting from Tom's Hardware.
