AI Data Center Profitability Hinges on Leasing and Financing Conditions

Domestic AI data center projects' success depends on more than just power acquisition, according to a Bitplanet Research Lab report released September 17. The analysis compared four domestic projects in Paju, Ulsan, Haenam, and Gumi, noting that even with confirmed power supply and demand, achieving profitability is not assured.

The report cited public data showing 522 applications for power system impact assessments at Seoul metropolitan area data centers from August 2024 to March 2026, totaling 33,592 megawatts (MW). Only 10 projects received final approval, amounting to 1,010 MW—a mere 1.9% of total applications. However, the report cautioned that ongoing reviews mean this percentage shouldn't be interpreted as individual project approval likelihood.

LG U+'s Paju AI data center is the only one with confirmed power supply (200MW) and pre-completion leasing for its first building. Exact lease terms remain undisclosed, making precise break-even point calculation difficult. SK's Ulsan facility has secured initial power but not final approval; Chairman Choi Tae-won announced 900MW expansion talks in September 11, though Korea Electric Power Corporation stated no formal procedures were submitted for additional supply.

Haenam National AI Computing Center, with a 40MW capacity and plans to host 15,000 AI chips, focuses on public infrastructure benefits alongside leasing revenue. Samsung SDS' Gumi data center, targeting 60MW, has announced a 427.3 billion won investment for construction by January 2026 but lacks confirmed site commencement or power approval as of September 15.

Bitplanet emphasized that comparing project investments is misleading due to varying business models and measurement standards. The report defined operational break-even as covering operating costs and loan interest—not accounting-based break-even or total investment recovery. For instance, LS Securities estimated a 25% average occupancy rate for Gwacheon IDC's break-even in July 2025, reflecting depreciation and interest burdens but not necessarily achievable outcomes.

Lease agreements don't immediately translate to revenue; Macquarie Infrastructure reported 99% of target IT load (25.44MW) was leased for Samseong IDC by Q1 2026, yet full rental income wasn't expected until mid-2027. Financing also plays a critical role: one project faced delayed land payment leading to loss of contractual benefits and early repayment demands.

Bitplanet Research Lab cautioned that public announcements about power contracts or tenant details alone don't determine financial success. The report stressed the need for transparency on actual lease terms, occupancy timelines, and post-operation revenue-cost data to accurately assess AI infrastructure viability.