A major new player has reportedly entered Korea's AI infrastructure arena.
Retail conglomerate Shinsegae is said to have reached a broad agreement with Reflection AI, a U.S. startup founded by former DeepMind researchers, to build a 250-megawatt (MW) AI data center. While still unconfirmed, this would be a landmark multi-billion dollar project. The deal structure is reported to be a partnership where Reflection AI provides the core technology—chips, AI models, and engineering—while Shinsegae handles financing, land acquisition, and permits. The facility is expected to house tens of thousands of Nvidia GPUs, placing it among the largest in the country.
This development didn't happen in a vacuum; it's the result of several converging factors. First, there's a strong national push. The Korean government is actively encouraging such projects, exemplified by its recent cooperation with BlackRock to develop hyperscale data centers. This policy support lowers the financial and regulatory hurdles for ambitious private builds. At the same time, domestic competition is fierce. With giants like Hyundai Motor Group and SK Telecom already investing billions in their own AI infrastructure, the pressure was on for other conglomerates to secure their own computing power or risk falling behind.
Second, the timing is right for the partnership. Reflection AI recently secured a substantial funding round of around $2 billion, giving it the financial stability to commit to large, long-term GPU purchases. High-level talks between the two companies were already happening in late 2025, signaling that this collaboration has been in the works for some time.
Finally, physical infrastructure constraints are shaping the landscape. The area around Seoul is facing a power shortage, pushing new, energy-intensive data centers to other provinces where the grid has more capacity. This project aligns with that trend. Furthermore, a major fire at a government data center in 2025 has increased the demand for more resilient and geographically dispersed facilities, making a new, state-of-the-art center an attractive proposition.
The project's scale is immense. A 250MW data center running at full capacity could consume about 2.19 TWh of electricity annually, which is nearly 0.4% of South Korea's total annual electricity consumption. This single project brings the critical issue of energy supply for the AI era into sharp focus, raising important questions about the future of the national power grid.
- Glossary:
- Hyperscale Data Center: A massive computing facility designed to efficiently support thousands of servers for large-scale applications, often run by major tech companies.
- Sovereign AI: A nation's ability to develop, deploy, and control its own artificial intelligence infrastructure and models, reducing dependence on foreign technology.
- PUE (Power Usage Effectiveness): A ratio that measures how efficiently a data center uses energy. The ideal PUE is 1.0, meaning all power goes to computing equipment.
