SK Telecom Builds Two Shells to Hold a Gigawatt It Hasn't Sold

The telecom repair is real and profitable; the AI data center is three years and three unnamed tenants away — and the dividend keeps sliding behind it

SKM · Earnings Call · 2026-11-04

The quarter that finally worked

After a year defined by an apology tour, SK Telecom's third quarter of 2026 reads like a company that has its feet back under it. Consolidated revenue came in at KRW 4,497.3 billion, up 3.2% quarter-on-quarter and 13.1% year-over-year, and operating income of KRW 552.3 billion marked the third consecutive quarter above the half-trillion mark. The CFO, Jong-seok Park, framed it as a promise kept: “With a cumulative consolidated operating income of KRW 1,655.9 billion for the first 3 quarters this year, we are continuing the trend of earnings recovery that we promised at the beginning of the year” — Jong-seok Park, CFO · 2026-11-04. The engine is unglamorous and that is the point. Mobile service revenue grew sequentially on the back of Galaxy Fold and iPhone launches, and management was explicit that the growth came from quality of subscribers, not quantity — high-LTV adds, foreigners as a new segment, and a deliberate refusal to re-enter the subsidy war that nearly broke the market. Notably, handset subscriber counts fell in the quarter while mobile service revenue rose. That is ARPU mix doing the work, and it is the cleanest evidence yet that the customer-trust repair after the 2025 breach is translating into money rather than just goodwill.

Two shells, one gigawatt, and somebody else's balance sheet

The real story of this call is structural, and it is told in two new entity names. SK Hyper, set up in July, holds the gigawatt-scale AI data center ambition. SK Horizon, announced in August and to be established early next year, holds the existing data center operations, a 100MW Ulsan build, and the submarine cable business. The logic given for splitting them is business stage: Horizon has contracts and a financing structure, so SK keeps a majority; Hyper needs to win customers and raise money, so it gets a looser framework. Read the second-order meaning. This is a company deliberately constructing the legal plumbing to carry external funding — project finance, strategic investors, financial investors — rather than its own equity. Management was candid about why: “AIDC business requires a massive amount of investment. So it is essential to secure stable demand through long-term contracts to be able to obtain external funding” — Unknown Executive, Executive · 2026-11-04. The sequence is demand first, capital second, concrete third — customer contracts targeted in the first half of next year, with the Ulsan facility opening in phases from the second half of 2029. That 2029 date is the tell. For all the AI infrastructure rhetoric, the revenue is three years out and contingent on three to four unnamed global counterparties signing. The CFO said it best when an analyst pressed on whether higher rates make the raise harder:

It is true that interest rates affect funding costs, but what matters more in securing investment is project bankability. As for SK Horizon, we were able to receive large external investments after we secured business stability and profitability based on long-term contracts.
Jong-seok Park, CFO · 2026-11-04

Fair — but bankability is exactly what has not yet been demonstrated at Hyper.

The market is de-rating the very thing SK is building

Here is the tension the transcript cannot see. Over the last 30 days, the global theme cluster around AI data center is in outright liquidation — roughly 11 advancers against 56 decliners across the ticker set, with high bandwidth memory, co-packaged optics and data center capacity all down double digits. SK Telecom is asking investors to underwrite gigawatt-scale AI capacity at precisely the moment the tape is repricing that asset class downward. That does not make the strategy wrong; it does make the "we will raise it externally" assumption harder, and it explains why the dividend question and the capex question have become the same question.

What management stopped saying

The silence is louder than the slides. Two quarters ago, in the May 2026 call, AI RAN and AI traffic were the two highest-momentum keywords in the entire transcript, with a network strategy executive walking analysts through base-station inference, NVIDIA and Samsung partnerships, and why AI traffic would reshape network economics. Both concepts have vanished from this call entirely. In their place: a growth gear strategy and a Token Gateway business that would meter customers' AI usage across models and infrastructure. It is a more commercial framing — and a less technically differentiated one. The pivot from "we will own the AI network" to "we will broker AI tokens" is a meaningful de-risking of the narrative. Second silence: profitability disclosure. In May, the CFO volunteered that “the AI DC business is comparable to the existing telecom business in terms of profitability” — Jong-seok Park, Chief Financial Officer · 2026-05-07. This quarter, no such claim was made. When the cost of building triples, the compare stops being flattering. Third, a genuinely new datapoint: standalone 5G now has a hard commercialization date of December 2026, after which 5G Advanced and 6G — standardization targeted for 2029, commercial launch after 2030 — follow. And a new worry surfaced from the sell side: low-earth-orbit satellite competition, following SpaceX's spectrum and base-station moves. Management's answer was dismissive and, to be fair, well argued: “the impact of satellite-based services on the domestic MNO market is expected to be quite limited” — Unknown Executive, Executive · 2026-11-04, given nationwide 5G and in-building coverage. But it is the first time this call has had to defend the core franchise from orbit.

The dividend that keeps slipping right

This is where the report gets uncomfortable. A year ago, after skipping a quarterly dividend, the same CFO said: “As a CFO, this is an extremely unsettling situation, and I'd like to extend my sincere apology to the investors” — Yang-Seob Kim, CFO · 2025-10-30. In February 2026 he promised to restore prior payout levels. In May 2026 he recommitted to “do our utmost to restore the dividend levels to the previous levels” — Jong-seok Park, Chief Financial Officer · 2026-05-07. Today the language is conditional and the timetable has moved: “we have just started the process of reviewing the future plan” — Jong-seok Park, CFO · 2026-11-04, with DPS set "as late as before the first quarter of 2027" — and explicitly conditioned on capex for AI data centers. Shareholders have now waited through three consecutive calls for a number that keeps being deferred behind the same project that has no signed customers. The Anthropic stake is the other half of that sentence, and its evolution is telling. In February the answer was that no decisions had been made. Now the CFO has a framework: proceeds will be weighed against the possibility of creating synergy between Anthropic and SK Telecom, market conditions, and necessary funds for growth investment. Translation: the stake is now explicitly on the table as AIDC funding. An IPO-linked windfall is being quietly re-designated from shareholder return to capital expenditure. The verdict. Operationally this is SK Telecom's best quarter in two years, and the telecom repair is real. Strategically it is a company admitting it cannot self-fund its AI ambition, so it is building separate vehicles to let others carry the risk — while the global market marks down exactly that asset class and the dividend waits on the outcome. The bull case needs three unnamed tenants to sign by mid-2027. The bear case needs nothing at all to happen.