June 8,SK Telecom and NVIDIA announce major collaboration, will build Gigawatt-scale AI cloud infrastructure in South Korea.Launched in 2027, will serve AI training, inference and agent workloads throughout South Korea and even the Asia-Pacific region.
Cooperation details: DSX full-stack architecture + gigawatt computing power
At the heart of this collaboration is NVIDIA’sDSX (full-stack AI factory reference architecture).Lowest single token cost, highest energy efficiencyTo support large-scale AI computing.
- scale: Gigawatt level (1 GW = electricity consumption of approximately 900,000 households)
- Architecture: NVIDIA DSX full-stack reference architecture
- First phase put into production: 2027
- Service scope: Training, inference, sovereign AI, physical AI, enterprise AI
Why is this important?
For readers who follow the AI industry, this cooperation has released several key signals:
First, the AI infrastructure race is expanding from the United States to Asia-Pacific.Previously, OpenAI’s Stargate project (450,000+ GB200 GPU) and Microsoft’s global data center plan were mainly concentrated in North America and the Middle East.
Second, the concept of “sovereign AI” has been officially implemented.NVIDIA and SK Telecom clearly mentioned that they will build "sovereign AI"-that is, autonomous AI infrastructure controlled by the government/domestic enterprises.
Third, the DSX architecture lowers the replication threshold for AI factories.NVIDIA packages hardware + software + operations into a standardized architecture, which means that any telecom operator with funds and power resources can quickly copy it.
What does it mean for the average user?
Although a gigawatt AI cloud sounds far away, its impact is real: more AI computing power meansFaster model response speed, cheaper API call costs, and more localized AI services.
Summarize
SK Telecom’s gigawatt-level cooperation with NVIDIA marks a new stage in the AI infrastructure race.AI DashWe will continue to track these changes.