The SK Group, fully mapped - the chaebol behind the AI memory supercycle

The standard way retail buys "the SK AI story" is to pick one ticker and assume it owns the memory. Almost every version of that assumption is wrong. The asset everyone actually wants, $HXSCL (SK hynix, the world's HBM leader), is a roughly $1.07 trillion company that no single accessible wrapper cleanly contains. It sits two holding-company layers deep inside a Korean conglomerate, its largest shareholder is a holdco most US investors have never heard of, and the one SK name everybody trades, $SKM, owns none of it.

This is the full map. Not which ticker to buy, that is the companion piece, but how the whole structure works: what a chaebol is, who owns whom, how value cascades up from the memory business and shrinks at every layer, and the two completely separate Anthropic stakes that retail keeps merging into one.

What a chaebol actually is

SK Group is South Korea's second-largest chaebol, after Samsung. A chaebol is a family-controlled conglomerate: dozens of legally separate listed and unlisted companies, bound together by a web of cross-shareholdings, steered from the top by a founding family that holds control far out of proportion to its economic ownership. SK's controlling family is the Chey family; the apex vehicle is SK Inc.

Two features of that structure drive everything in this article.

First, control and economics are deliberately decoupled. The family controls the group through a chain of holding companies, each owning enough of the next to direct it, without owning most of the underlying cash flows. A ~30% stake one layer down, repeated across layers, is plenty to steer a trillion-dollar asset the family economically owns a single-digit slice of.

Second, every holding layer historically trades at a discount. When a listed company's main asset is shares in another listed company, the market rarely pays full value for the wrapper. It applies a holding-company discount: for tax friction on the embedded gains, for the family control that subordinates minority holders, and for capital that gets allocated across the whole group rather than to the shareholders of any single piece. In Korea those discounts have run especially deep, which is the second story in this piece.

The family tree, top to bottom

The simplified ownership map, top down:

• SK Inc. (KRX 034730) is the group holding company, the apex the Chey family controls. It holds the controlling stakes in the listed subsidiaries below it, roughly 30% of SK Square and roughly 30% of SK Telecom among them.
• SK Square (KRX 402340) is the technology-investment holdco, spun out of SK Telecom in November 2021. Its defining asset is a ~20% stake in SK hynix, which makes it the chipmaker's single largest shareholder. Smaller stakes in commerce, mobility and platform businesses sit around that core.
• SK Telecom ($SKM) is Korea's largest mobile carrier, ~45% subscriber share. Since the 2021 spin-off it is a sibling of SK Square, not a parent of anything semiconductor.
• SK hynix (KRX 000660 / HXSCL) is the crown jewel: ~60% of the HBM market, lead HBM3E and HBM4 supplier to $NVDA, and the anchor of QA's DRAM / HBM Memory bubble.
• The supporting cast: SK Innovation (energy and EV batteries via SK On), SKC (semiconductor materials and glass substrates via Absolics), SK Ecoplant (construction and, increasingly, datacenter builds).

The shape to hold in your head: SK Inc. on top, SK Telecom and SK Square as siblings beneath it, and SK hynix hanging off SK Square. The value everyone wants sits at the bottom of that chain, and you reach it by climbing down through the discounts.

Why SK Telecom split in two in 2021

The single fact that breaks most retail SK theses dates to November 2021. SK Telecom, then one company that both ran the telecom network and held the hynix stake, split itself in two:

• SK Telecom kept the operating telecom business: the network, the subscribers, the dividend.
• SK Square was spun out as a separate, listed investment holdco, and the SK hynix stake went with it.

The logic was a Korean version of a sum-of-the-parts unlock. A regulated, slow-growth, high-dividend telecom and a volatile, high-beta semiconductor holding were dragging on each other inside one share price. Splitting them let income investors own the carrier and growth investors own the chip exposure, and it gave SK Square a clean mandate as the group's technology-investment arm.

The consequence that matters five years later: SK Telecom has not owned a single share of SK hynix since that day. Every pitch that says "buy SKM for the hynix exposure" is describing a company that stopped existing in 2021.

The crown jewel, and why everything else is a derivative of it

SK hynix is covered in depth in its own explainer and in HBM is the tightest bottleneck in the AI cycle, so the short version: it leads the highest-pricing-power corner of memory, HBM, sold on multi-year contracted capacity to NVIDIA and the hyperscalers. At the June 12, 2026 close it carried a market cap of about ₩1,523 trillion (~$1.07T), the second-largest company on the KOSPI behind only Samsung Electronics (~₩2,052T). It out-earns and out-margins $MU while the memory cycle runs hot.

Hold that ₩1,523 trillion number. Every other SK vehicle in this article is, to a first approximation, a fractional and discounted claim on it. The rest of the map is just accounting for how big the fraction is and how deep the discount.

The money cascade: how value flows up, and shrinks at each layer

Here is the whole structure priced at the June 12, 2026 close, top to bottom, with the look-through arithmetic that makes the discounts visible. Won converted at roughly ₩1,420 per dollar and rounded.

| Layer | Ticker | Market cap | What it holds | | --- | --- | --- | --- | | SK Inc. | KRX 034730 | ₩32.3T (~$23B) | ~30% of SK Square, ~30% of SK Telecom, plus SK Innovation, SKC, SK Biopharm | | SK Telecom | SKM | ₩21.9T (~$15.4B) | The telecom, plus a direct ~0.3% Anthropic stake | | SK Square | KRX 402340 | ₩178.8T (~$126B) | ~20% of SK hynix | | SK hynix | KRX 000660 / HXSCL | ₩1,523T (~$1.07T) | The memory business itself |

Now walk it from the bottom.

Hynix to SK Square. SK Square owns 20.07% of SK hynix. At a ₩1,523T hynix cap, that stake is worth about ₩306 trillion. Yet SK Square's own market cap is ₩178.8T. The market values the wrapper at roughly 0.58x the value of the single asset inside it, a ~42% discount, before you even count SK Square's smaller commerce and mobility holdings. That gap is the holding-company discount, made concrete.

SK Square to SK Inc. SK Inc. owns about 30% of SK Square. At SK Square's ₩178.8T cap, that stake alone is worth about ₩54 trillion, already 1.7x SK Inc.'s entire ₩32.3T market cap, and that is before adding SK Inc.'s ~30% of SK Telecom, its controlling stake in SK Innovation, and the rest. SK Inc. trades below the market value of just one of its holdings. That is a discount applied on top of a discount: the double-decker version of the same Korean holdco phenomenon.

The cascade, in one line: a ₩1,523 trillion asset, seen through a ~42%-discounted holdco, seen again through a parent that trades at little more than half its stake in that holdco. Value flows up the chain, and a piece of it evaporates into discount at every step.

The two discounts, and the reform closing them

A holding-company discount is the gap between a holdco's market cap and its net asset value (NAV): the summed market value of everything it owns, minus its debt. SK Square's NAV is dominated by hynix, and its discount is the ~42% gap above. This is not a Korean oddity in kind, only in degree. Naspers traded for years at a deep discount to its Tencent stake; Softbank trades below the sum of its holdings; the structure invites it.

What is specific to Korea right now is the Value-Up program, the government-backed corporate-reform push modeled on Japan's, pressuring listed companies to close their price-to-book gaps through buybacks, dividends, and governance changes. It gave the market a reason to re-rate the wrappers themselves, not just the assets underneath.

The effect on SK Square has been the standout chart of the Korean market. Its NAV discount narrowed from 65.7% at the end of 2024 to 51.5% at the end of 2025 to roughly 42-45% by spring 2026. The stock is up about 192% year to date, crossed the symbolic million-won mark, and is now the #3 market cap on the KOSPI, behind only Samsung Electronics and SK hynix itself. SK Square has leaned in: its first-ever cash dividend (~₩200B, structured tax-exempt) plus ~₩110B of buybacks in 2026, funded partly by the ~₩710B of dividends it collects from hynix, with a stated goal of trading above 0.7x NAV by 2028.

Two things follow, and they cut in opposite directions. When the discount narrows, the wrapper outperforms the asset: you earn hynix's move plus the re-rate, which is exactly what the last 18 months were. When the discount is done narrowing, the wrapper becomes hynix-with-extra-steps: the same memory-cycle downside, plus governance and conglomerate risk, minus the liquidity. The discount is the trade, in both directions.

SK Inc.'s discount is the same phenomenon one floor up, and structurally deeper, because it stacks the family-control and capital-allocation concerns of the apex vehicle on top of SK Square's. It is the cheapest look-through claim on the whole complex, and the messiest.

The Anthropic confusion, fully untangled

This is where retail, and a fair amount of professional, coverage goes wrong. There are two completely separate Anthropic stakes inside the SK complex, held by two different companies, for two different reasons. They are not the same investment, and neither one is held by SK Square.

Stake one: SK Telecom's direct equity. SK Telecom put $100M into Anthropic in August 2023 (following an earlier, smaller check), at a ~$5B valuation, through its own corporate venture arm, SK Telecom Venture Capital (SKTVC), the Santa-Clara-based fund that had first taken SKT into Anthropic's Series C. Because SKTVC is SK Telecom's subsidiary, the stake sits on SK Telecom's balance sheet, carried at roughly ₩1.3T (~$970M) at end-2025. Diluted by every mega-round since, it is now about 0.3% of Anthropic, down from ~0.7% at entry. SKT made a further follow-on in the $965B Series H that closed in late May 2026, and its CEO confirmed in Tokyo on June 10 the intent to hold through Anthropic's IPO. This is the "backdoor Anthropic" stake, and at the Series H mark Morningstar describes it as dominating SK Telecom's ~$15B valuation.

Stake two: SK hynix's strategic chip-supply stake. Separately, SK hynix took its own stake in Anthropic as part of that same Series H, alongside Samsung and Micron, as a strategic infrastructure partner. The individual amounts were never disclosed and are, by every indication, small: this is a supply-chain handshake, early sight into Anthropic's future memory specs and a foot in the door on next-generation HBM, not a financial position. For hynix, the value is selling the memory, not owning the equity.

Now the trap. SK Square holds neither. Its only link to Anthropic is two hops removed: SK Square owns ~20% of SK hynix, and SK hynix owns a tiny strategic stake. That is why Anthropic never appears in the SK Square NAV breakdowns analysts publish. So:

• Want the direct, disclosed, quantified Anthropic equity stake? That is SK Telecom, through SKTVC.
• Want exposure to the company that supplies Anthropic's memory? That is SK hynix, and at a discount, SK Square.
• The two get merged constantly because SK Square is "the SK investment company" and an Anthropic stake feels like it belongs there. It does not. The fund that bought Anthropic is SK Telecom's, not SK Square's.

The full proxy math on the SKT stake, alongside $AMZN, $GOOGL and the funds, is in How to invest in Anthropic.

The four tickers as exposure, side by side

Map the structure onto what you can actually own and the four core tickers resolve into four different bets. The full trade-selection treatment, with access mechanics, is the companion piece, one chaebol, three ways to own the AI memory leader. The short version:

• SK hynix (HXSCL) is the asset itself. Cleanest exposure, full memory-cycle beta, thin US access until the SEC-registered ADR targeted for end-2026 lists. Live data: /stocks/hxscl.
• SK Square (402340) is discounted, leveraged hynix. The closing-NAV-discount trade, Seoul-listed only, with ~40% of the gap already closed after a 192% run.
• SK Inc. (034730) is the deepest-discount, look-through claim on everything, with the most governance and capital-allocation noise stacked on top.
• SK Telecom (SKM) is the only one with a clean US listing, a 3.7% dividend, and the direct Anthropic stake. It is an Anthropic proxy wearing a carrier's clothes, and it owns no hynix. Live data: /stocks/skm.

The thesis on the underlying memory cycle barely changes across the four. The wrapper you pick changes the trade far more than it changes the thesis: which discount you are exposed to, which currency, which liquidity, and whether you are really buying memory or buying Anthropic. Two of the four are realistically Seoul-only for US retail; /stack/ibkr covers which account setups can reach the KRX.

The supporting cast

Three group companies earn a mention in an AI-supercycle map, as optionality rather than the core thesis:

• SKC / Absolics: SKC's Absolics subsidiary is building glass-core substrates for advanced chip packaging in Covington, Georgia, with US CHIPS Act support. Glass substrates are a credible next step for the packaging layer that HBM-heavy accelerators are straining. Early and capex-heavy.
• SK Innovation / SK On: EV batteries and energy. Central to the group's balance sheet and its capital-raising decisions, mostly irrelevant to the memory thesis except as a reminder that the chaebol allocates capital across the whole family, not just the jewel.
• SK Ecoplant: construction moving into datacenter builds, a small, indirect tie to the same AI-infrastructure demand the memory business sells into.

The bull case

• The asset is real. Roughly 60% of HBM, the lead supplier on NVIDIA's current and next memory generation, in the tightest-supply corner of the AI stack.
• Two re-rating engines, not one. The hynix earnings cycle plus the Value-Up discount compression. SK Square's 18-month chart is what it looks like when both fire together.
• The access unlock is dated. A liquid US hynix ADR targeted for end-2026 widens the buyer base for the whole complex and gives every wrapper a cleaner reference price.
• The discount still exists. Even after the rally, SK Square's NAV gap of roughly 40%-plus leaves room if Value-Up pressure persists; Japan's analogous reform cycle ran for years, not quarters.
• An IPO that marks the hidden asset. An Anthropic listing converts SK Telecom's dominant unlisted stake into a marked, sellable position.

The bear case

• Memory is still a cycle. Every layer of this tree is ultimately a claim on DRAM/NAND/HBM pricing. When the cycle turns, the holdco discount historically widens at exactly the wrong moment: leverage in both directions.
• Chaebol governance is the discount's reason. Cross-shareholdings, family control, and capital allocation that serves the group before minority shareholders. Value-Up is policy pressure, not a law of nature.
• Dilution overhang. The hynix US listing is backed by a large primary raise; new supply near the listing can cap the complex.
• FX and access friction. KRW exposure on every vehicle except SKM, and two of the four tickers are realistically Seoul-only for US retail.
• The Anthropic mark is unrealized. Until the IPO, SKM's dominant AI asset is a private valuation, not a sellable position, and the narrative-fit flows that bought it can leave just as fast.

What to watch

• The hynix US ADR terms: final size, the new-share component, the symbol, and timing into end-2026. The single biggest event for every wrapper in the tree.
• SK Square's NAV discount: whether it holds in the 40s, keeps compressing on Value-Up momentum, or re-widens on a memory wobble. The discount is the trade.
• Value-Up policy cadence: index-inclusion rules, tax incentives, and whether holdco reform survives Korea's political cycle.
• HBM4 share retention: hynix versus Micron and Samsung on the next NVIDIA generation, which sets the asset's earnings path.
• The Anthropic IPO (targeted H2 2026): the event that converts SKM's dominant unlisted asset into a marked position.
• SKM earnings 2026-08-05: how much of the re-rate the actual telecom numbers can support, and the scale of the AI-datacenter segment against the narrative.

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

---

Live data on these tickers: /stocks/hxscl and /stocks/skm - price, bubble correlation, related research.

Bubble context: /bubbles/memory - the DRAM / HBM Memory cluster the crown jewel anchors.

Adjacent reading: one chaebol, three ways to own the AI memory leader, SK hynix is heading to a US listing, and How to invest in Anthropic.

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