KRW 3.08 trillion. That is the cheque SK Telecom just raised for a company that does not exist yet.
The ownership split is 51% for SK Telecom, 29% for KKR, 20% for the Korean consortium, with the telecom keeping control while selling 49% of its AI data center platform to institutional capital.
The new company inherits eight operating data centers plus two AI data centers under construction, targets 318 MW of capacity, and pairs compute with submarine cable infrastructure under one roof.
Telecom operators are the quiet owners of the AI infrastructure boom. They hold the land, the power contracts, the fibre, and the construction permits that hyperscalers need. The question has always been who pays for the buildout. SK Telecom's answer, announced on August 27, is the most explicit yet: a separate company, priced and funded like infrastructure, with outside equity on the cap table.
Institutional capital in an AI data center platform
KKR and the IMM Investment-Stonebridge consortium commit KRW 3.08 trillion for a combined 49% of the spun-off vehicle. SK Telecom, August 2026
The carve-out behind the cheque
The new company is the second piece of a three-part structure SK Telecom has been assembling since July. SK Telecom itself holds strategy and relationships with global technology partners. SK Hyper, created in July, develops gigawatt-scale AI data centers, aiming for 5 GW by 2029 and 15 GW by 2035. The new entity now takes the operating layer: eight live data centers in Seocho, Ilsan (two sites), Bundang, Gasan, Centum, Yangju, and Pangyo, plus the AI data centers under construction in Ulsan and Guro.
The spin-off splits the wholly owned subsidiary SK Broadband on a book-value ratio of roughly 0.84 to 0.16. The surviving SK Broadband keeps fixed-line, media, and enterprise services. Everything attached to data center operations moves into the new company, including the submarine cable programme. That last point matters more than it looks: international connectivity is the hidden input of frontier AI training, and most data center deals never bundle the pipe with the building.
Kim Seong-soo, who runs SK Broadband, is expected to serve concurrently as the new company's CEO. The spin-off is scheduled to complete in the first quarter of 2027, pending shareholder and government approvals.
"This governance restructuring is a proactive measure aimed at strengthening expertise and enabling faster execution in the AIDC business."— Kim Seong-soo, CEO, SK Broadband
What KKR and the Korean consortium are actually buying
Read the cap table and the deal stops looking like a funding round. It looks like a structuring event. SK Telecom retains 51%, keeps management control, and continues to set strategy. KKR gets 29% through its Asia Pacific infrastructure strategy, which sits inside a firm with more than $100 billion in infrastructure assets under management and over $70 billion across digital and power assets. The IMM Investment-Stonebridge consortium takes 20%, pairing a global infrastructure manager with domestic Korean institutional capital: IMM, founded in 1999 with over $7.5 billion under management, and Stonebridge Capital, which manages roughly KRW 3.6 trillion.
The buyer side reveals the thesis. KKR is not buying GPUs or a customer list. It is buying a regulated utility-like asset: land, power, fibre, and long-duration contracts, with a growth option attached. The 318 MW target across eight plus two facilities is the current operating envelope. The growth option is SK Hyper's 5 GW and 15 GW pipeline, which stays outside the platform but feeds it. Infrastructure funds buy the annuity; the upside lives in the pipeline.
That is why the submarine cable sits in SK Horizon rather than SK Broadband. In most telecom structures, cable landing stations and international fibre are bookkept alongside consumer broadband. Placing them in the data center company tells the market these are AI infrastructure assets, valued at AI infrastructure multiples, not telecom multiples.
Why bundle data centers with submarine cables?
The pattern underneath
This is the third major signal this year that AI data center capacity is being repriced as core infrastructure. First came the hyperscaler lease deals and private credit funds targeting compute. Then KKR's own Helix Digital Infrastructure platform, seeded with more than $10 billion in June, alongside the Kuwait Investment Authority and Nvidia. Now a listed telecom takes the same step inside its own asset base.
The mechanics are consistent. A capital-intensive asset gets separated from the operating company that cannot fund it cheaply enough. Institutional capital takes a minority or majority stake. The operating partner keeps control and the buildout gets funded. The same template appears in cell towers, fibre networks, and now AI data centers. As we wrote in August, when we mapped where value actually sits in the AI infrastructure stack, the owners of capacity and power earn the structural returns. SK Telecom did not invent the structure. It applied it to the most strategically important asset it owns.
The Korean angle adds another layer. Sovereign and pension capital across Asia has been rotating into digital infrastructure, and the government has declared AI data centers one of its three national mega-projects. A domestic consortium holding 20% alongside KKR makes the transaction easier to approve and politically palatable, which matters when the split needs regulatory sign-off.
The risks are the usual ones for infrastructure announced before it is built. Power supply remains the binding constraint on Korean data center expansion. The 318 MW target is a plan, not a contract. And the Q1 2027 completion date sits behind shareholder approval and government review. None of that changes the direction of travel: the capital for AI infrastructure is now structurally separate from the operators who build it.
What happens to SK Horizon a year from now?
Probability: 70%, because the ownership structure is fixed, the assets are operating today, and the precedent is cheap for a rival carrier to copy.
✅ Arguments for
Confirmation criteria: first external customer announcement for Ulsan or Guro within two quarters; 318 MW milestone dates in regulatory filings.
❌ Arguments against
Disconfirmation criteria: no power purchase agreement for Ulsan by mid-2027; SK Telecom quiet on SK Horizon in two consecutive earnings calls.
Power procurement: does the new company sign its own PPAs, or stay inside SK Telecom's contracts?
Anchor tenants: hyperscaler or sovereign AI commitments for Ulsan and Guro capacity.
Cable landing deals: which subsea routes the new company takes on, and with which partners.
Copycat spin-offs: whether KDDI, Singtel, or another Asian carrier restructures its data centers the same way.
Development scenarios
🟢 Optimistic scenario (20%)
Implications: SK Telecom's AI infrastructure business is valued on infrastructure multiples, unlocking capital for SK Hyper's 15 GW pipeline.
🟡 Base-case scenario (55%)
Implications: A stable, moderately growing infrastructure asset. Institutional investors earn infrastructure-grade returns; the telecom keeps control of strategic compute capacity.
🔴 Pessimistic scenario (25%)
Implications: KKR and IMM hold a slower asset; the structural template survives but loses its pricing power for rival spin-offs.