SK Telecom (SKM): not the SK Hynix backdoor everyone thinks — what you're actually buying

The standard $SKM pitch is that it's a cheap backdoor into SK Hynix: buy the sleepy Korean telecom at a discount, get the HBM-memory giant in the wrapper for free. It's the single most repeated reason retail buys the ADR.

It's also wrong. SK Telecom owns zero SK Hynix. The ~20.5% stake was spun off to SK Square in November 2021 — four years ago. What you're actually buying in $SKM is a discounted Korean carrier wrapped around two very different AI assets: a pre-IPO stake in Anthropic marked at Anthropic's ~$965B round, and a gigawatt-scale AI-datacenter buildout with AWS. Neither is memory.

What SK Telecom actually does

The base business is boring and durable: SK Telecom is South Korea's #1 mobile operator, roughly 45% market share, ~24M subscribers, a best-in-class nationwide 5G network, plus fixed-line broadband (via SK Broadband) and enterprise services. It throws off stable, high-margin free cash flow and pays a dividend yielding ~3.7%.

On top of that base, management has bolted an AI strategy it calls the "AI Pyramid" — three layers:

1. AI Infrastructure — datacenters, GPU-as-a-Service, AI semiconductors, and its own LLM.
2. AIX (AI transformation) — applying AI to the telco and selling it to enterprises; it co-founded the Global Telco AI Alliance (with Deutsche Telekom, e&, Singtel, SoftBank) to build a shared "Telco LLM."
3. AI Service — consumer agents: Adot (Korea, with a Perplexity search partnership) and Aster, a global "agentic" personal assistant.

The pitch is a telecom re-rating itself into an AI-infrastructure company. The market is only paying for one piece of that — and it's not the piece most buyers think.

The SK Hynix myth

This is the correction that reframes the whole name. The "free SK Hynix" thesis is a common retail misattribution, and the timeline is unambiguous:

• In November 2021, the old SK Telecom split into SK Telecom (the carrier, what trades as $SKM) and SK Square (an investment holdco).
• The ~20.5% SK Hynix stake went to SK Square, not to SK Telecom.

So if your thesis is "cheap way to own the HBM memory leader," the vehicle is SK Square (which trades at a wide NAV discount to its Hynix stake) — or SK Hynix directly ($HXSCL, which is prepping a US listing). It is not SK Telecom. For how the chaebol is split across entities, see the SK Group structure breakdown.

The real AI asset: the Anthropic stake

SK Telecom's actual frontier-AI exposure is an Anthropic position:

• A $100M check in August 2023, topped up in the $965B Series H that closed May 2026.
• Morningstar has reframed the entire equity as "becoming an Anthropic proxy," because the mark on that stake now dominates the valuation of a company with a market cap around $14.6B.
• Management has signaled hold-through-IPO intent (Anthropic's listing is targeted for H2 2026, possibly October), and the relationship is deepening operationally: SKT joined Anthropic's invite-only "Project Glasswing" in June 2026 with early access to a gated Claude security model, and Anthropic needs compute that SKT intends to supply.

Here's the mechanic that matters for valuation, and the reason the multiple looks strange: the Anthropic gain sits in Other Comprehensive Income (equity), not net income. It inflates book value, not EPS. So the stake doesn't make the P/E look cheap — if anything trailing P/E looks high because core telecom earnings are soft. The discount, where it exists, is a price-to-book / sum-of-parts story, not a low-earnings-multiple one. Treat any unusually low printed "P/E" with suspicion: a one-off gain (see the datacenter stake sale below) can distort trailing earnings for a quarter or two.

The buildout: Ulsan and a gigawatt of AI datacenter

The second real AI asset is physical. SK is building South Korea's largest AI datacenter in Ulsan, with AWS as anchor on a 15-year agreement:

• Total capex ~₩7T (~$5.1B), including roughly $4B from AWS.
• Power ramp: 41 MW by Nov 2027 → 103 MW by Feb 2029, with a stated long-term target of 1 GW.
• ~60,000 GPUs initially; ultra-high-density racks on hybrid air + liquid cooling.

And the capital structure is the tell: SK Telecom is selling up to 49% of the project (KKR ~29%, plus an IMM–Stonebridge consortium) at a ₩2–2.5T ($1.3–1.9B) valuation, with the agreement targeted for end of June 2026. That's the institutional asset-light playbook — de-risk the capex, keep operating, recycle capital into the broader AI push. It's the same "who funds the megawatts" pattern reshaping the whole datacenter-power bubble.

Around it sits the rest of the infrastructure layer: GPU-as-a-Service ("Petasus AI Cloud," with a Lambda partnership), an in-house LLM (A.X), and an AI-inference chip effort via Rebellions (with Arm, aimed at sovereign AI). A gigawatt-scale AI Cloud with $NVDA silicon is targeted for 2027.

The numbers

• Market cap: ~$14.6B (the ADR is small relative to how the name gets discussed).
• Valuation: trades at roughly ~6x EV/EBITDA vs ~7.5–9x for Western telco peers — the classic "Korea discount." Forward P/E is in the mid-teens; trailing is distorted by depressed earnings.
• Dividend: ~3.7% yield, supported by the mature carrier cash flows.
• Re-rating tailwind: Korea's Corporate Value-Up program is pushing Korean companies toward better shareholder returns and higher valuations.
• Data as of: 2026-06 (QA /stocks/skm panel).

The bull case

• Anthropic proxy at an early vintage, marked at the $965B Series H, with hold-through-IPO intent and a concrete catalyst (Anthropic IPO targeted H2 2026).
• A real, contracted buildout — AWS's 15-year anchor makes the Ulsan datacenter contracted offtake, not a speculative build; the KKR stake sale validates the asset value and frees capital.
• Carrier cash flows + a ~3.7% dividend fund the AI pivot without forcing dilution.
• Korea Value-Up re-rating plus deepening Anthropic operational ties (Project Glasswing, compute supply).

The bear case

• The Anthropic mark is unrealized until the IPO. A down-round or a delay re-rates the entire narrative, because the stake is what's carrying the valuation.
• Core telecom is ex-growth — revenue roughly flat, earnings down year-on-year, with price-cap and saturation pressure in Korea.
• The valuation has already run on the proxy story — this is not the cheap, unnoticed name it's sometimes pitched as.
• Wrapper + governance friction — KRW/ADR access, a persistent chaebol/holdco discount, and the 2025 USIM data-breach overhang (subscriber churn and remediation cost).

How to access

• $SKM is the NYSE ADR (USD) — the simplest US-account route. 017670 is the ordinary line on the Korea Exchange (KRW).
• ⚠️ Ticker trap: the US symbol SKT is Tanger Inc., an outlet-mall REIT — completely unrelated. Don't type SKT expecting SK Telecom; use SKM. To trade the ADR from a US-retail account, see /stack/ibkr.
• If your actual thesis is memory, that's SK Hynix ($HXSCL) or SK Square, not this name.

Bubble-correlation shifts and rule-based alerts on $SKM are part of /pro.

What to watch

• Anthropic IPO (H2 2026, possibly October) — the single biggest catalyst; it converts the unrealized mark into a realized, priced number.
• The Ulsan 49% stake-sale close (~end June 2026) — confirms the $1.3–1.9B asset valuation and books a one-off gain that will distort trailing earnings.
• Ulsan power ramp — 103 MW committed vs the 1 GW option; watch for new PPAs and the AWS offtake schedule.
• Korea Value-Up follow-through — buybacks/payout changes would attack the holdco/Korea discount directly.
• The read that would change: if the market ever starts pricing SKM as if it owns SK Hynix again, that's the misattribution returning — the correct frame stays Anthropic + datacenter, not memory.

---

Live data on this ticker: /stocks/skm — price, ETF holdings, bubble correlation, bot positions.

Bubble context: /bubbles/datacenter-power — the cluster this name belongs to and how it's moving.

QuantAbundancia is educational research. Nothing here is investment advice. See /disclosures.