Fifty-five megawatts is the opening number. NAVER and NVIDIA announced a sovereign AI factory partnership, per the companies’ announcement coverage on NVIDIA’s blog, which confirmed active South Korea sovereign AI infrastructure content including a June 4–7 post on how NVIDIA and South Korea are building AI infrastructure. The specific partnership article URL is unconfirmed, a Source Hint is used here for the direct announcement.
The reported structure: NVIDIA’s DSX platform provides the full infrastructure stack at NAVER’s GAK Sejong facility. The 55MW initial phase is the first step in what NAVER and NVIDIA describe as long-term plans to scale to gigawatt capacity, a forward-looking figure that, like all multi-year infrastructure projections, should be treated as aspirational until capital commitments are confirmed.
The model development layer is where this gets interesting for enterprise buyers. NAVER is developing what it calls a “Seoul World Model” under this sovereign AI initiative, alongside next-generation HyperCLOVA X, NAVER’s established AI model brand, verifiable from prior industry reporting. The framing here is explicitly nationalist: a Korean-language-first, Korea-hosted, NVIDIA-powered AI stack that doesn’t run through US hyperscaler infrastructure for inference or training.
That’s the pattern. This announcement follows a now-recognizable template: national AI champion plus NVIDIA full-stack infrastructure plus a regionally-branded model. The same structure appeared in France (SoftBank/Sesterce sovereign AI factory), in the UK, and now in South Korea. NVIDIA confirmed UK sovereign AI coverage dated June 7, 2026, in the same blog content batch.
The catch is that the sovereign AI factory framing can obscure what’s actually being bought. When NAVER licenses NVIDIA DSX and builds NVIDIA-powered infrastructure, the “sovereign” element refers to data residency, regulatory compliance, and government ownership of the resulting model, not to independence from US technology supply chains. The compute stack is still NVIDIA. The software platform is still NVIDIA. What Korea owns is the model weights trained on that infrastructure and the data that stays within Korean jurisdiction.
Analysis
This is the third sovereign AI factory announcement following a consistent NVIDIA DSX + national champion + regional model template in recent cycles. The pattern suggests NVIDIA has standardized its go-to-market for national AI infrastructure deals. Enterprise buyers evaluating sovereign vs. hyperscaler AI infrastructure should compare on data residency requirements, language support, and regulatory compliance needs, not on hardware differentiation, which is minimal.
For enterprise teams evaluating AI infrastructure in East Asia, this matters in a specific way. If your workload requires Korean regulatory compliance, Korean-language performance at scale, or Korean data residency, the NAVER/HyperCLOVA X stack is worth evaluating. If your requirement is simply “best performance per dollar,” the sovereign AI label doesn’t change the underlying infrastructure comparison.
Don’t confuse the “sovereign AI” brand with a genuinely different technology architecture. The underlying hardware and platform are still NVIDIA. What’s sovereign is the governance, the data, and the model, and for the use cases where those things matter, that’s exactly what’s needed.