What's Happening?
SK Telecom has announced an ambitious plan to develop up to 15 gigawatts (GW) of AI data center capacity across South Korea by 2035. The initiative will begin with the construction of a hyperscale site in Ulsan, in partnership with AWS, expected to start
operations in the latter half of next year. The project is part of a broader national strategy to establish South Korea as a leading hub for AI infrastructure. The company plans to bring 5 GW online in phased stages starting in 2029. This development is part of a larger government initiative targeting 18.4 GW of AI data center capacity by 2035, involving major partners like SK Group, GS Group, and Naver. The estimated cost for a 1 GW-class AI data center is approximately 70 trillion won, with funding sources including self-investment, strategic partners, and project financing.
Why It's Important?
This expansion is significant as it positions South Korea as a major player in the global AI infrastructure landscape. The development of such large-scale AI data centers will create substantial demand for high-bandwidth memory, specialized power, and cooling systems, and will likely stimulate local supply chains. For machine learning teams, the availability of onshore hyperscale capacity can reduce latency and data egress costs, although it may not fully address scheduling or provisioning challenges due to constrained accelerator supply. The project also aligns with South Korea's national strategy to secure core AI elements faster than other countries, potentially giving it a competitive edge in AI technology development.
What's Next?
The next steps involve SK Telecom reviewing power, site selection, operating systems, and key customers as part of the rollout. The company will likely structure capacity access through owned racks, colocation, managed services, or committed co-location contracts. Observers should watch for procurement windows, long-term contracting, and how SK Telecom structures these agreements. Additionally, the staging of the Ulsan hyperscale site, contract terms offered to enterprise customers, and any new grid capacity projects will be critical indicators of when onshore capacity becomes available for large-model training or persistent inference fleets.
