Can China's and Hong Kong's AI Overheat Spill Into Korea?

Sun, 31 May 2026 15:30:00 +0900

Context This note follows our work on Upstage and Korean sovereign AI, SK Telecom as an AI infrastructure operator, and AI token futures and cost per token. The question here is narrower: if Hong Kong and China are willing to price scarce LLM listings aggressively, where can that narrative land in Korea’s listed market? Related hubs: Korean AI Companies Hub and Korea Stocks for Foreign Investors Hub.

TL;DR

Yes, the Hong Kong / China AI heat can spill into Korea. But it is unlikely to arrive as a clean Zhipu- or MiniMax-style listed LLM pure-play trade. Korea almost does not have such a stock.

The spillover is more likely to happen through four proxies.

Spillover path	Korea proxy	View
Sovereign LLM scarcity	NAVER, private Upstage	Possible, but NAVER is a platform with an LLM option, not an LLM pure-play
AI cloud / enterprise AI	LG CNS, Samsung SDS, SK AX	The most realistic route, because revenue and customers are visible
AI data center / GPUaaS	SK Telecom, parts of NAVER Cloud	Direct, but utilization and capex payback must be proven
LLM venture proxy	SK Networks, parts of SK Inc.	Event-driven exposure to Upstage valuation / IPO optionality

The key conclusion:

If China’s and Hong Kong’s AI overheat transmits to Korea, it will probably not be a “buy the Korean MiniMax” trade. It will be a reclassification trade in companies that can turn AI usage into recurring revenue, margin, and capex payback.

The cleanest listed candidates are LG CNS and SK Telecom. LG CNS has the clearest Cloud&AI revenue mix. SK Telecom is beginning to show real AIDC / GPUaaS numbers. Samsung SDS has the strongest narrative, but the numbers need to catch up. NAVER owns the largest sovereign AI option, but it is not a pure-play. SK Networks is an Upstage event proxy.

1. Why Hong Kong and China matter

The core signal from Hong Kong and China is simple.

The market is willing to price scarce foundation-model listings as sovereign AI call options, not as normal near-term earnings stories.

Zhipu / Knowledge Atlas is one of the clearest examples. It owns the GLM model family and monetizes through API and private deployment. Its 2025 revenue was roughly RMB724 million, up 131.9% year on year, but net loss was about RMB4.72 billion. The stock is therefore not about current earnings; it is about scarcity, domestic model capability, and the “China OpenAI” option.

MiniMax follows a similar logic. Its 2025 revenue was reported at about US$79 million, up 158.9% year on year, while net loss remained large. Investors are paying for a global AI app, multimodal model optionality, and LLM scarcity rather than present profitability.

Enterprise AI and AI cloud names are valued differently. Fourth Paradigm, SenseTime, Baidu, Alibaba, and Tencent all sit in different baskets. The market asks whether AI revenue is recurring, whether API usage is scaling, whether gross margin after inference cost is attractive, and whether capex will dilute the economics.

Basket	Multiple logic	Korea translation
LLM / foundation model pure-play	Extreme P/S can appear	Almost absent in listed Korea
AI app / consumer AI	High growth premium possible	Private Upstage, some NAVER service optionality
Enterprise AI platform	Lower P/S or EV/EBIT	LG CNS, Samsung SDS, SK AX
AI cloud / big tech	Normal PER plus capex discount	NAVER, Samsung SDS, parts of SKT
AI infra / GPUaaS	EV/EBITDA and capex payback	SK Telecom, parts of NAVER Cloud

The practical question is therefore not “Does Korea have a listed Zhipu?” It does not. The better question is: which Korean listed companies can borrow part of the China/Hong Kong AI valuation language?

2. Korea’s missing slot: no listed LLM pure-play

Korea has almost no direct Zhipu / MiniMax equivalent in the public market.

NAVER is the only large listed Korean platform with a meaningful proprietary LLM option: HyperCLOVA X, CLOVA Studio, NAVER Cloud, AI search, and AI-driven ad / commerce optimization. But NAVER’s revenue mix is still search, ads, commerce, fintech, content, and C2C. NAVER reported Q1 2026 revenue of KRW3.2411 trillion and operating profit of KRW541.8 billion, and it reclassified revenue into NAVER Platform, Financial Platform, and Global Initiatives. (NAVER)

That makes NAVER a platform company with an LLM option, not an LLM company.

The more realistic public-market route is AI cloud integration. LG CNS and Samsung SDS do not sell foundation models as their core product. They help enterprises deploy AI through cloud migration, MSP, GPUaaS, agent workflows, public-sector AX, and managed platforms.

Category	Listed in Korea?	Representative names	View
LLM pure-play	Almost no	Private Upstage is closest	No direct listed comp
Large listed proprietary LLM owner	Yes	NAVER	Sovereign LLM optionality
AI cloud integrator	Yes	LG CNS, Samsung SDS, SK AX	The most realistic spillover path
AI data center / GPUaaS	Yes	SK Telecom, parts of NAVER Cloud	Asset-heavy AI infra proxy
AI venture / LLM proxy	Yes	SK Networks	Indirect Upstage exposure

3. Company-by-company translation

3.1 NAVER: listed sovereign LLM owner, not a pure-play

NAVER has Korea’s most visible listed LLM option. HyperCLOVA X, CLOVA Studio, NAVER Cloud, AI Search, advertising, and commerce all connect to the thesis. NAVER said AI accounted for more than 50% of advertising revenue growth in Q1 2026, and it also referenced sovereign AI opportunities. (NAVER)

But the right equation is not “LLM revenue x LLM P/S”.

NAVER AI value
= better defense and monetization in search / ads / commerce
+ Enterprise AI / NAVER Cloud revenue
+ sovereign AI optionality
- AI capex and compute cost burden

NAVER can receive a scarcity premium if Korea investors start looking for listed sovereign AI exposure. But the pace of re-rating depends on AI revenue disclosure and margin defense.

3.2 Samsung SDS: KKR-backed full-stack AI integrator, but numbers need to catch up

Samsung SDS is not a model company. It is an AI cloud integrator with a strong narrative: FabriX, Brity Copilot, Brity Automation, Samsung Cloud Platform, GPUaaS, public-sector AX, and KKR-backed capital.

In Q1 2026, Samsung SDS cloud revenue was KRW690.9 billion, up 5.8% year on year. Against KRW1.6105 trillion of IT services revenue, cloud accounted for 42.9%.

Cloud revenue share = KRW690.9bn / KRW1.6105tn = 42.9%

Samsung SDS reported Q1 revenue of KRW3.3529 trillion and operating profit of KRW78.3 billion. Operating profit fell sharply due to a one-time KRW112 billion retirement-benefit provision. The company also laid out a KRW10 trillion investment plan by 2031, including KRW5 trillion for AI infrastructure. (Samsung SDS)

Samsung SDS can attract the strongest AI full-stack narrative in Korea. The constraint is proof: cloud growth, AI SaaS revenue, GPUaaS margin, recurring revenue, and public AX conversion must show up in the numbers.

3.3 LG CNS: the cleanest listed Korean AI cloud integrator

LG CNS is not an LLM company either. But it may be the cleanest listed AI cloud integration proxy.

The Elec reported that LG CNS posted Q1 2026 revenue of KRW1.315 trillion and operating profit of KRW94.2 billion. Its core AI and cloud segment revenue reached KRW765.4 billion, or about 58% of total sales. (The Elec)

Cloud&AI revenue share = KRW765.4bn / KRW1.315tn = 58.2%

Its portfolio includes public-sector, defense, finance, manufacturing, pharma, biotech, and shipbuilding AX projects; global cloud partnerships; OpenAI and Palantir relationships; agentic AI workflows; and data-center DBO.

The upside is not a Zhipu-style LLM multiple. The upside is that LG CNS may deserve a higher multiple than a traditional SI if Cloud&AI revenue is recurring, high-margin, and locked into customer workflows.

3.4 SK Telecom: the most direct AI infrastructure proxy

SK Telecom can be reclassified from telco to AI data center / GPUaaS / sovereign AI infrastructure operator.

In Q1 2026, SKT reported consolidated revenue of KRW4.3923 trillion, operating income of KRW537.6 billion, and net income of KRW316.4 billion. Its AIDC business recorded KRW131.4 billion of revenue, up 89.3% year on year, helped by higher data-center utilization and GPUaaS revenue. (SK Telecom)

Annualized AIDC revenue = KRW131.4bn x 4 = KRW525.6bn
Annualized group revenue = KRW4.3923tn x 4 = KRW17.5692tn
AIDC share = KRW525.6bn / KRW17.5692tn = 3.0%

The share is still small. But it is real and growing. SKT combines AIDC, GPUaaS, the A.X sovereign model, the A. agent, B2B AI, and group synergies with SK hynix, SK ecoplant, and SK Innovation / E&S.

SKT is not CoreWeave. But in the Korean listed market, it is one of the most direct AI infrastructure operators.

3.5 SK Inc.: AI full-stack holding company, not AI software multiple

SK Inc. is not an LLM stock or cloud integrator. It is an AI full-stack holding company.

SK Square and SK hynix provide AI memory exposure. SK Telecom provides AIDC / GPUaaS. SK AX provides enterprise AX. SK ecoplant and materials subsidiaries provide infrastructure and materials. Energy and cooling sit in the wider group. SK Inc.’s Q1 2026 briefing also shows SK hynix HBM / eSSD strength and SK Telecom AIDC / GPUaaS contribution. (SK Inc.)

The right framework is NAV discount, not LLM P/S.

SK Inc. appeal
= AI-related affiliate NAV growth x holding discount compression
- non-AI asset, debt, and overlap risk

3.6 SK Networks: listed Upstage call option

SK Networks is not a fundamental AI operating company. It is a public-market proxy for Upstage optionality.

Maeil Business reported that SK Networks, an existing Upstage investor, is known to have invested an additional KRW50 billion in Upstage’s Series C round. Upstage raised KRW180 billion in the first close and was valued above KRW1 trillion. (MK) VentureSquare separately reported the same KRW180 billion first close and unicorn valuation. (VentureSquare)

What remains unclear is important: exact SK Networks ownership, post-money dilution, fair-value accounting, and IPO timing.

SK Networks is therefore an event proxy for Upstage valuation / IPO optionality, not a core AI operating-company re-rating.

3.7 Upstage: private benchmark for Korean LLM valuation

Upstage is the closest Korean private LLM pure-play. Solar, Document AI, private deployment, and sovereign AI exposure make it the key benchmark.

The public market will care about the following metrics if Upstage moves toward an IPO.

KPI	Why it matters
Model / API revenue	Core LLM pure-play metric
Document AI ARR	Real enterprise monetization
On-prem / private deployment revenue	Regulated sovereign AI demand
Gross margin after inference cost	True LLM economics
Inference cost per 1M tokens	Model efficiency
Customer ARR / NRR / GRR	Proof that it is recurring software, not SI
Overseas revenue share	Global AI app / service optionality

The better framing is not “Korea’s OpenAI”. It is an LLM-native workflow platform for regulated enterprise document and knowledge work.

4. Ranking the Korea proxies

AI cloud / enterprise integration

Rank	Company	Why
1	LG CNS	Cloud&AI is already about 58% of revenue; cleanest listed proxy
2	Samsung SDS	Strong KKR / FabriX / Brity / GPUaaS narrative; needs proof
3	SK AX / SK Inc.	Relevant, but small within SK Inc. NAV

AI infrastructure / GPU cloud

Rank	Company	Why
1	SK Telecom	AIDC revenue +89.3% YoY; GPUaaS demand is visible
2	NAVER	NAVER Cloud, GPUaaS, sovereign AI; smaller within group revenue
3	Samsung SDS	Some SCP / GPUaaS exposure; more integrator than infra pure-play

LLM / sovereign AI

Rank	Company	Why
1	Upstage	Private, but closest Korean LLM pure-play
2	NAVER	Only large listed proprietary LLM platform
3	SK Telecom	A.X exists, but LLM revenue disclosure is limited

Event / proxy

Rank	Company	Why
1	SK Networks	Upstage IPO / valuation proxy
2	SK Inc.	SK Group full-stack AI NAV proxy
3	Samsung SDS	KKR / AI infrastructure investment events

5. Failure conditions

The AI spillover can happen, but it can also fail quickly if the market sees only labels without numbers.

Theme	Spillover probability	Failure condition
LLM pure-play P/S	Low	No listed Korean pure-play
Sovereign LLM scarcity	Medium	Weak NAVER AI revenue disclosure; delayed Upstage IPO
AI cloud integrator	High	AI revenue proves to be low-margin project SI
AI data center / GPUaaS	High	Utilization, margin, and capex payback stay undisclosed
Upstage proxy	Event-driven	Ownership / fair value remains opaque; IPO delayed

The question is not “Does the company do AI?”

The better question is: Can the company convert AI usage into revenue, recurrence, margin, and capex payback?

Conclusion

The Hong Kong / China AI heat can spill into Korea, but the form will differ. Korea lacks a listed Zhipu or MiniMax. The local trade is therefore a reclassification trade in existing listed companies.

LG CNS and SK Telecom are the cleanest structural candidates. Samsung SDS has the strongest narrative but needs numbers. NAVER has the largest listed sovereign AI option but is not a pure-play. SK Networks is an Upstage call option.

One-line version:

In Korean AI stocks, the key is not “who is the LLM?” It is who can translate rising AI usage into recurring revenue, margin, and capex recovery?

SK Inc on Korea Invest Insights