Nvidia and SK Group are building 2GW of Korean AI capacity
Nvidia and SK Group are planning up to 2 gigawatts of AI data-center capacity in Korea. That is not just another GPU order. SK is combining Nvidia compute, SK hynix memory, telecom infrastructure, construction, and energy into one stack, with the first AI factory planned for 2027.
Nvidia President & CEO Jensen Huang tells Bloomberg's Ed Ludlow “This is the golden ages for Korea" after teaming up with SK Group to build more than 2 gigawatts of AI data centers on the Korean Peninsula. Watch the full interview: https://t.co/8vxJ9Ul3zZ pic.twitter.com/Uw3EV3SHNJ
— Bloomberg TV (@BloombergTV) July 25, 2026
Q1What was officially announced?
Nvidia and SK Group announced a wider partnership covering AI factories and next-generation memory. SK Telecom plans to build an Nvidia-powered AI cloud of up to 2 gigawatts in Korea, using Vera Rubin systems, the DSX architecture, and SK hynix HBM4 memory. The first AI factory is planned for 2027.
Q2How big is two gigawatts?
It is enormous for one AI infrastructure program. Two gigawatts equals the continuous power demand of twenty 100-megawatt data centers. It is also twice the scale of a 1-gigawatt campus, a threshold that only the largest AI projects are now starting to target. This puts Korea inside the global gigawatt compute race.
Q3Was SK already building with Nvidia?
Yes, but the scale has jumped. In October 2025, SK and Nvidia announced an AI factory with more than 50,000 GPUs, with its first phase planned for late 2027. SK also partnered with AWS on an AI data center in Ulsan. The new 2-gigawatt target turns those separate projects into something closer to national-scale infrastructure.
Q4Why is SK unusually well positioned?
Because SK controls several parts of the stack. SK hynix makes the high-bandwidth memory Nvidia needs. SK Telecom can sell the compute as a cloud service. Other SK companies bring data-center construction, networks, gas, and energy infrastructure. Most AI builders need many outside partners to assemble the same pieces.
Q5What is the real point of tension?
The bottleneck is moving from buying GPUs to powering and operating them. A 2-gigawatt plan needs huge grid connections, cooling systems, land, financing, and a reliable supply of advanced memory. Nvidia can provide the computing platform, but Korea still has to turn a very large announcement into working electrical and industrial infrastructure.
Q6Why does this matter for Korea?
Korea is already central to AI because Samsung and SK hynix supply memory chips. This project tries to move the country further up the stack, from selling components to operating the machines that train and run AI. If the capacity reaches global customers, Korea could become a major exporter of compute as well as chips.
Q7So should I believe the full 2GW?
Treat it as a serious target, not finished capacity. The first factory is planned for 2027, while the full build will require years of construction and power development. The next proof points are actual sites, secured electricity, rack deployments, paying cloud customers, and how quickly SK moves from its first phase toward the full two gigawatts.
