NAVER's 55-megawatt start — a $10bn sovereign AI factory rises in Sejong
NAVER raised $10bn from NVIDIA and Brookfield to build an AI factory starting at 55MW in 2027, targeting 1GW long-term
The three lines
- NAVER builds an NVIDIA DSX-based AI factory at its Sejong hyperscale campus, live at 55MW in H1 2027
- $10bn (~₩14.6tn) from Brookfield and NVIDIA lifts the initial buildout to 200MW
- The long-term 1GW target makes this Korea's sovereign-AI infrastructure flagship
Key questions
- What exactly is NAVER building
- An AI factory — a data center purpose-built for training and running AI, based on NVIDIA's DSX platform, at NAVER's GAK Sejong hyperscale campus. The roadmap is staircase-shaped: 55 megawatts live in the first half of 2027, 100MW later that year, 200MW in 2028, and a long-term target of one full gigawatt.
- Where is the money coming from
- A $10 billion (about ₩14.6 trillion) raise from Brookfield, the Canadian alternative-asset manager, and NVIDIA itself. The investment lifted the initial buildout from 55MW to 200MW. It is an unprecedented infrastructure raise for a Korean internet company — and notable for NVIDIA entering as an investor, not just a supplier.
- Why does the gigawatt number matter
- Scale context: Korea's all-time peak power demand is 97.1GW — this single facility ultimately plans to draw about 1% of national peak load. AI competition is shifting from chips to electricity and land, which is why the project sits in Sejong, and why it belongs in the same conversation as this week's power-grid records.
The center of gravity in AI competition is moving from models to megawatts. NAVER — Korea's dominant search and internet company — has raised $10 billion from NVIDIA and Brookfield to build an "AI factory" in Sejong: live at 55 megawatts in early 2027, scaling toward a full gigawatt, the power draw of a nuclear reactor. In the week this page covered China's five-models-in-eight-weeks barrage, Korea's answer arrives not as a model but as a factory.
1. What gets built — a factory in the literal sense
The AI-factory metaphor is precise: electricity and data go in, GPU racks are the machinery, and inference — answers — comes out. NAVER's design runs on NVIDIA's DSX data-center platform inside GAK Sejong, its hyperscale campus. The roadmap climbs in steps: 55MW in H1 2027, 100MW by year-end, 200MW in 2028, 1GW as the long-term target.
The money explains the project's real nature. Brookfield — an infrastructure asset manager — and NVIDIA itself put in $10 billion, lifting the initial buildout from 55MW to 200MW. Two signals there: the GPU monopolist is now investing in its customers' capacity, and global infrastructure capital has started treating AI data centers as power-plant-grade assets.
2. The roadmap, with scale context
| Stage | Timing | Capacity | Context |
|---|---|---|---|
| Go-live | H1 2027 | 55MW | Mid-size data center |
| Expansion 1 | H2 2027 | 100MW | |
| Expansion 2 | 2028 | 200MW | Raised by the investment |
| Long-term | TBD | 1GW | ~1 reactor's output |
| Investment | Brookfield + NVIDIA | $10bn (~₩14.6tn) | |
| Site | GAK Sejong | — | Power and land drove the choice |
The gigawatt figure belongs on the same screen as this week's other Korean power story. National peak demand just hit 95.3GW, 1.8GW from the all-time record — and this one facility ultimately plans to draw about 1% of national peak. AI infrastructure competition has become a grid-planning problem, which is exactly what "sovereign AI" means in practice: not a model, but land, power, chips and capital assembled under national control.
3. What remains
Two watch items. The model: whether NAVER's own AI can justify factory-scale infrastructure — distribution without a competitive model is an empty highway. And the grid: how a 1GW private load gets written into national power planning. What "sovereign AI" precisely means, and who else is building one, is today's companion reference, "Sovereign AI explained"; the global model race context runs in "Five models in eight weeks."