Kakao

Kakaocorp.com

South Korea’s Kakao Corp. approved the largest governance overhaul in its 16-year history on Friday, voting to divide itself into two separately listed companies — KakaoAI, built around the country’s dominant messaging platform, and KakaoX, an investment-focused holding arm for its fintech, mobility, and entertainment businesses. Investors responded with a sharp sell-off: Kakao shares fell to ₩34,150 won (approximately $24.71 USD) by mid-morning on the Korea Exchange, down nearly 12% by mid-session, erasing gains built on months of AI momentum and record quarterly earnings.

The board’s calculation is blunt: Kakao’s sprawling conglomerate structure has been destroying shareholder value at scale. Domestic and international securities analysts peg its sum-of-the-parts value at 34.2 trillion won (approximately $24.75 billion USD), against a market capitalization hovering near 16.8 trillion won (approximately $12.16 billion USD) — a discount of more than 50%. The new thesis is structural: separate the businesses, give each independent governance and capital allocation, and let investors price each entity on its own growth merits rather than penalizing the whole for the complexity of its parts.

That thesis had not yet convinced the market by mid-session Friday.

Under the approved demerger plan, KakaoAI will be spun off as a new publicly listed company housing KakaoTalk — South Korea’s dominant messaging platform, used by approximately 94.7% of the total population and 97.2% of all South Korean internet users — alongside Kakao’s AI, advertising, and commerce businesses. The surviving entity, rebranded as KakaoX, will retain the group’s fintech arm (including Kakao Pay and Kakao Bank), the Kakao T mobility platform, and its entertainment and content affiliates.

The split ratio, set at 0.36 for KakaoAI and 0.64 for KakaoX based on each entity’s net asset book value, means existing shareholders will receive stakes in both companies proportional to their current holdings. Kakao stated the move was aimed at giving businesses with different characteristics and stages of growth greater autonomy in strategy, execution, and capital allocation.

“This split is a decision to transform Kakao into a structure built for the speed and accountability required in the AI era,” said CEO Chung Shin-a, who will lead KakaoAI. KakaoX will be headed by Kim Do-young, current CEO of Kakao Investment, who told reporters at a Friday briefing that there are “absolutely no plans” to restructure KakaoX as a holding company.

Kakao founder Kim Beom-su, who holds approximately a 24% stake in Kakao Corp. and will remain the largest shareholder of both entities after the split, framed the announcement in strategic rather than financial terms: “In the mobile era, Kakao was among the first to venture into uncharted territory, creating innovations that changed users’ daily lives. The AI era demands agility on an entirely different level. We will redesign our growth structure around the two engines of Kakao AI and Kakao X so that the value they create can be shared with shareholders, users and our crew.”

Kakao plans to seek shareholder approval for the demerger at an extraordinary general meeting on December 17, 2026, with the split scheduled to take legal effect on January 1, 2027. KakaoAI and KakaoX shares are scheduled to begin trading separately on January 27, 2027.

Why Markets Sold Off What Management Called a Discount Fix

The 12% stock decline invites a question the board’s announcement does not fully answer: if the sum of the parts is worth twice the current market price, why aren’t investors buying?

The answer lies in the difference between theoretical value and realized value. A sum-of-the-parts analysis assigns each business unit the valuation multiple that comparable standalone companies trade at — Kakao Pay benchmarked against fintech peers, KakaoBank against digital banking peers, KakaoTalk against messaging platform peers. The aggregate exceeds the current market price because the market applies a discount for holding-company complexity, cross-subsidiary accountability gaps, and the governance overhang from Kakao’s recent legal and regulatory history.

Separation, in theory, eliminates the complexity discount. But it introduces new uncertainty: how will each entity trade on its own? What multiple will investors apply to KakaoAI, a profitable AI-pivoting messaging business with a 2030 revenue target of at least 6 trillion won (approximately $4.34 billion USD)? What discount will KakaoX trade at as an investment-focused conglomerate holding fintech and mobility assets — the very structure critics fault as the source of Kakao’s current discount?

As one analysis put it: “once core businesses and subsidiary stakes are separated into different listed companies, the appropriate value and growth potential applied to each entity may be reassessed, creating short-term uncertainty.” The market’s Friday reaction suggested that “short-term uncertainty” is currently outweighing “long-term value unlock,” according to analysts covering the stock.

Investor sentiment has also been shaped by Kakao’s track record. The company’s aggressive affiliate expansion in the early 2020s attracted regulatory scrutiny for anti-competitive practices and drew criticism for what Korean financial media described as an “octopus-like” conglomerate structure. The company’s record quarterly earnings — ₩277 billion (approximately $200 million USD) operating profit in Q2 2026, up 36% year-over-year — have demonstrated genuine improvement. But the legal shadow over its founder has not lifted.

What Is KakaoAI’s Actual Bet?

Stripping away the corporate restructuring mechanics, KakaoAI is a concentrated bet on one claim: that KakaoTalk, with its near-total Korean market penetration, is the most powerful possible distribution infrastructure for an AI agent that executes transactions on users’ behalf.

KakaoAI’s vision is to evolve KakaoTalk into an “agentic AI interface” that understands the intent and context of each of its approximately 50 million users, with personalized AI agents connecting discovery, recommendations, purchases, and payments within conversations. The company has set a 2030 target of more than 20 million daily active AI users, a 50% increase in platform dwell time, and consolidated revenue of at least 6 trillion won (approximately $4.34 billion USD).

Central to that strategy is Kanana, Kakao’s proprietary AI model family. The company has been preparing to unveil Kanana 2.5, a 150-billion-parameter model developed from scratch for agentic AI deployment. At 150 billion parameters, Kanana 2.5 occupies less than 10% of the parameter count of frontier global models like GPT-4-class systems — but Kakao CEO Chung said it outperforms larger models in the task categories KakaoAI actually needs: planning and function calls, the two capabilities an AI agent executing commerce transactions must excel at.

How Kanana’s Tokenizer Unlocks the Economic Math

The technical mechanism that makes the KakaoAI economics viable — and explains why Kakao is investing in a proprietary model rather than licensing a Western frontier model — is the Kanana tokenizer.

Language models do not process text the way humans read it. They divide text into units called tokens, which the model then processes. The problem for Korean is fundamental: standard tokenizers, optimized for English, require 1.5 to three times more tokens to express the same Korean meaning as an equivalent English sentence. More tokens mean more compute per inference. More compute per inference means higher operating costs at scale. For a platform aiming to serve 50 million users with conversational AI at a cost point competitive with advertising revenue, this arithmetic is not peripheral — it determines whether the business model works.

Kakao’s custom Kanana tokenizer was built to solve this. According to CEO Chung’s Q1 2026 earnings call disclosure, the Kanana tokenizer achieves the strongest Korean-language compression efficiency among disclosed Korean and overseas models. The claimed gains: up to 40% training cost reduction and up to 60% improvement in inference speed, compared with standard tokenizers. At 50 million users, a 60% inference speed improvement is the difference between an AI agent that feels instant and one that breaks the habit formation Kakao is trying to engineer into daily commerce behavior.

Below Kanana 2.5 in the model stack sits Kanana-2-1.3B — a 1.3-billion-parameter small language model designed to run directly on users’ mobile devices rather than requiring a cloud server round-trip for each interaction. Kakao open-sourced the model on Hugging Face, making it auditable by external developers. On-device inference eliminates per-query server costs, enables privacy-preserving processing by keeping conversational context local, and reduces the latency that would otherwise break agentic commerce flows. The tradeoff: on-device models have hard parameter limits and require domain specialization that cloud-scale models avoid. Kakao’s bet is that Korean-language commerce specialization is sufficient.

These two models — Kanana-2-1.3B for on-device triage and Kanana 2.5 for cloud-side orchestration — form the technical spine of KakaoAI’s consumer proposition.

What Is KakaoX?

KakaoX is not merely a residual holding structure for assets KakaoAI didn’t want. The company has been positioned as a future-value investment company overseeing major business lines including fintech, content, and mobility, as well as new growth businesses. KakaoX has set its own 2030 revenue target of at least 10 trillion won (approximately $7.23 billion USD).

Separately, Kakao announced a 300 billion won (approximately $217 million USD) share buyback and cancellation plan, funded by proceeds from the sale of its stake in Dunamu, as part of broader shareholder return measures.

KakaoX CEO nominee Kim Do-young’s denial that KakaoX will become a holding company is significant because it addresses market concern that splitting off KakaoAI leaves a holding company shell — the least attractive corporate structure in Korean financial markets. The company said the two entities will operate under independent management without the joint CA Council that previously coordinated affiliate operations across the group.

What Kim Beom-su’s Ongoing Appeal Means for the Split

One element of the restructuring that the board announcement did not address — but that investors and legal analysts are watching closely — is the status of the criminal appeal against founder Kim Beom-su.

Kim was arrested in July 2024 on charges of manipulating SM Entertainment’s stock price during Kakao’s 2023 acquisition to block rival HYBE’s competing bid. Seoul Southern District Court acquitted him of all charges in October 2025. Prosecutors appealed, citing omitted evidence assessment. The appellate trial at Seoul High Court began in early 2026; the next hearing is scheduled for August 26, 2026, with a ruling expected in October 2026.

The timing creates a specific governance risk for the split’s December 17 shareholder vote. South Korean financial law restricts those convicted of financial crimes from holding more than a 10% stake in a bank. Kim Beom-su controls approximately 24% of Kakao Corp. today; post-split, he will hold equivalent stakes in both KakaoAI and KakaoX. KakaoX is the entity retaining Kakao’s ownership of KakaoBank. If Kim is convicted on appeal before or shortly after the shareholder vote, that ownership structure could become subject to regulatory action under South Korean banking law.

Kakao has not addressed this contingency publicly. The appellate ruling’s expected October 2026 timing means it will land approximately six weeks before the December 17 extraordinary general meeting — potentially making the founder’s legal status a material variable in how major institutional shareholders vote on the split.

Conglomerate Discount: Korea’s Persistent Problem

Kakao is not the first Korean technology and internet conglomerate to diagnose its own valuation gap and attempt a structural remedy. The “Korean discount” — a broader phenomenon in which South Korean listed companies trade at lower valuations than comparable global peers — has been attributed to opaque governance, cross-holdings, controlling shareholder structures, and limited minority shareholder protections.

Kim Beom-su stated the valuation gap directly at Friday’s briefing: “We will eliminate the conglomerate discount that undervalues the company by more than 50%.” That the gap he is targeting is structural rather than performance-based is notable: Kakao’s operational trajectory has improved markedly, with three consecutive quarters of operating profit growth heading into 2026. The discount persists not because the business is performing poorly but because the market does not trust a holding structure with cross-subsidiary governance to optimize capital allocation across diverse assets.

The board’s fundamental calculation — that two focused, independently governed companies will be valued more than one sprawling holding company — has precedent globally. But it also has cautionary examples. Not every demerger produces the anticipated unlock. The test here will come in January 2027, when KakaoAI and KakaoX begin trading separately and investors must independently price two companies rather than one.

Markets Await the December Test

Between now and December 17, Kakao’s management faces the task of converting institutional skeptics into shareholder supporters. Friday’s 12% decline suggests the initial reaction has been risk-averse: investors who bought Kakao as a diversified Korean internet play may not want two narrower exposures, and investors who want pure-play KakaoAI exposure may prefer to wait for the new shares to trade independently in January 2027 rather than entering a position in the combined company now.

Analysts covering the stock remained bullish heading into the announcement: the consensus 12-month target price stood at approximately ₩76,593 won (~$55.42 USD), representing a potential upside of 78.8% from pre-announcement levels and over 124% from Friday’s post-announcement price of ₩34,150 won (~$24.71 USD). Whether that gap between analyst models and market price narrows before December 17 will depend on whether Kakao can demonstrate, over the next four months, that the two separate businesses are worth pricing independently — not at a combined discount.

The December shareholder vote will be the next key test. For now, Friday’s board approval marks the formal starting gun for what could become a defining restructuring in the history of South Korea’s internet economy.

Frequently Asked QuestionsWhat is the Kakao split ratio, and what does it mean for shareholders?

The split ratio is 0.36 for KakaoAI and 0.64 for KakaoX, based on the net asset book values of each entity’s assets. Existing shareholders will receive shares in both companies proportional to their current holdings — meaning if you hold 100 Kakao shares today, after the split you will hold shares in both KakaoAI and KakaoX, no dilution occurs, and no cash changes hands. The practical question is how each entity will be priced independently once they begin trading separately on January 27, 2027. KakaoAI, housing KakaoTalk and the AI/advertising/commerce businesses, will likely attract a higher growth multiple. KakaoX, housing fintech, mobility, and entertainment investments, may trade more like a diversified investment holding company.

Why did Kakao stock fall 12% when the company said the split would increase shareholder value?

The gap between a structural value-unlock thesis and an immediate market reaction reflects two different time horizons. Analysts’ sum-of-the-parts model suggests the businesses are worth more separately than together — hence the 50%-plus gap between the 34.2 trillion won (~$24.75 billion USD) analyst valuation and the approximately 16.8 trillion won (~$12.16 billion USD) market capitalization. But markets price uncertainty, not theoretical value, and the split introduces several unknowns: how each entity will trade independently, whether KakaoX’s structure will attract its own valuation discount, and whether the ongoing appellate trial against founder Kim Beom-su — with a ruling expected in October 2026 — creates a governance risk ahead of the December 17 shareholder vote.

What technical capabilities will KakaoAI bring to KakaoTalk, and how does Kanana work?

KakaoAI’s core technical asset is the Kanana model family, which powers KakaoTalk’s AI features. The model family’s technical architecture and specifications has two main layers: Kanana-2-1.3B, a small language model that runs directly on users’ mobile devices for low-latency conversational AI, and the forthcoming Kanana 2.5 at 150 billion parameters for cloud-side agentic orchestration. Kanana 2.5 uses a custom Korean-language tokenizer that Kakao says cuts training costs by up to 40% and improves inference speed by up to 60% compared with standard tokenizers — addressing a fundamental bottleneck where Korean requires 1.5 to three times more tokens than English to express the same meaning. Above both models sits an orchestrator-agent architecture: a lightweight controller that delegates tasks to specialized agents, each capable of completing real-world actions (map queries, restaurant reservations, Kakao Pay payments) without per-step user confirmation.

What happens if Kim Beom-su is convicted on appeal before the December 17 shareholder vote?

South Korean financial law restricts those convicted of financial crimes from holding more than a 10% stake in a bank. Kim Beom-su controls approximately 24% of Kakao Corp. today and will hold equivalent stakes in both KakaoAI and KakaoX after the split. KakaoX will retain Kakao’s ownership position in KakaoBank. The Seoul High Court’s appellate ruling is expected in October 2026 — approximately six weeks before the December 17 extraordinary general meeting. A conviction on appeal could trigger regulatory review of Kim’s KakaoBank-adjacent ownership through KakaoX, creating governance uncertainty at precisely the moment institutional shareholders are being asked to approve the restructuring. Kakao has not addressed this contingency in its public communications.