Upstage just bought Daum from Kakao — the first time a Korean LLM company owns a search portal outright
On May 7, Korean LLM startup Upstage closed the acquisition of AXZ, the operator of Korea's Daum portal, from Kakao. Structured as a share swap: Kakao hands over 100% of AXZ and takes a stake in Upstage in exchange. Upstage will rebuild Daum as a 'context AI portal' with its Solar LLM and Daum's index and content corpus. The KFTC merger review is the last variable.
A Korean LLM company finally owns a domestic search portal
On May 7, Korean LLM startup Upstage closed the acquisition of AXZ — the operator of Kakao's Daum portal — from Kakao Corp. Not a cash sale. A share swap: Kakao transfers 100% of AXZ to Upstage and takes shares of Upstage in return. Upstage owns Daum; Kakao becomes a major shareholder of Upstage.
Two reasons this matters. First — the first time a Korean LLM company owns a search portal outright. OpenAI built SearchGPT, Anthropic built search into Claude.ai, Perplexity created the new "answer engine" category. In Korea, neither Naver nor Kakao is a native LLM company, and LLM startups (Upstage, Konan, Saltlux) had no proprietary search index. This deal closes that gap on the Korean side.
Second — Kakao's strategic retreat. KoGPT was an early start, but Kakao lost ground in the LLM race. Daum's search share (~4–5%) is dwarfed by Naver (~60%+), making it effectively a stranded asset. Trading Daum to Upstage for shares creates more option value than running it directly. Kakao keeps focus on KakaoTalk (messenger), KakaoPay (payments), KakaoMobility, and KakaoEntertainment, while gaining LLM/search exposure through Upstage equity.
The cast — Upstage, Daum (AXZ), Kakao, Solar LLM, Naver
Upstage. Founded in 2020 by Sung Hoon Kim (ex-Naver Clova) and three co-founders. Korean LLM company building the Solar series. Closed Series B at $80M in 2025; estimated valuation ~$400M. Post-Daum-deal, valuation is plausibly $800M–1B. Small globally, but with Daum it becomes the largest Korean / Southeast Asian / Japanese player owning a vertically integrated index + LLM stack.
Daum (AXZ). Founded 1995, one of Korea's first-generation web portals. Merged with Kakao in 2014. The Daum business unit was later spun off as a Kakao subsidiary, AXZ. As of 2026: ~4–5% search share, ~20M MAU, hundreds of millions of daily page views — material assets, but eclipsed by Naver. Daum Mail, Daum Cafe, and Daum News still hold significant content positions.
Kakao. ~25 trillion KRW market cap (May 2026). Korea's super-app: KakaoTalk (50M MAU), KakaoPay, KakaoT. KoGPT effort started in 2022 but trails LG AI Research's EXAONE, Naver's HyperCLOVA, and Upstage's Solar. Strategic retreat from search and LLM into the Upstage cap table.
Solar LLM. Upstage's flagship product. Solar 1 (2023), Solar 2 (2024), Solar 3 mini/pro/ultra (2025–2026). Claims +15–20% vs. same-size Llama/Mistral on Korean benchmarks. Top-2 in Korean-language category on global LLM leaderboards. With Daum's index and corpus, Solar 4 plausibly approaches GPT-5/Claude class on Korean reasoning and generation.
Naver. ~60%+ Korean search share. Owns HyperCLOVA X, Cue: search chatbot, Clova Note. Until now, the only player in Korea with both an LLM and a search stack. Now sees a real competitor in the form of Upstage + Daum.
Korean government. MSIT supports the deal under "sovereign AI" policy. The Korea Fair Trade Commission (KFTC) is reviewing for competitive impact in search/portal/LLM markets. Combined Upstage + Daum search share is under 5%, so KFTC clearance is the likely outcome.
What's inside — deal mechanics, Solar + Daum synergy, "Context AI Portal"
Deal structure. Kakao transfers 100% of AXZ to Upstage. In exchange, Upstage issues new shares to Kakao. The exact equity percentage isn't disclosed; industry estimates put it at 25–35% — making Kakao Upstage's #2 or #3 shareholder. Post-deal cap structure roughly: founders/employees ~35%, existing VCs ~30–35%, Kakao ~25–35%, Daum balance sheet ~5%. Operational control stays with founders + VC syndicate.
Solar + Daum synergy. What Upstage gains:
- Korean search index — Daum's 30 years of accumulated Korean web index. Direct fuel for Solar's RAG.
- Content corpus — Daum Cafe, Daum News, Daum blog content. Training data (subject to licensing).
- User base — 20M MAU as instant distribution for an "AI Daum" launch.
- Ad infrastructure — Daum's ad system, paired with LLM responses, opens experiments in "AI advertising" formats.
Context AI Portal vision. Upstage's published direction for Daum: shift from keyword search to intent-and-context-aware answers. A query like "cafe in Seoul on a rainy day" no longer returns keyword matches but a Solar-generated answer integrating user location, time, past preferences, and current weather. Korean-language Perplexity / SearchGPT — but with proprietary index + proprietary LLM, no US model dependency.
| Item | Detail |
|---|---|
| Closing date | 2026-05-07 |
| Structure | Share swap (no cash) |
| Estimated Kakao stake in Upstage | ~25–35% |
| Daum search share (2026) | ~4–5% |
| Daum MAU | ~20M |
| Upstage post-deal valuation (est.) | $800M–1B |
| Solar LLM Korean benchmark rank | Top 1–2 in size class |
| KFTC review | Ongoing |
| Upstage Series B (2025) | $80M |
Copyright and data. Using Daum Cafe/blog/news content for LLM training requires user consent + publisher licensing. Korean Copyright Act + PIPA + AI Basic Act (effective Jan 2026) all apply. The biggest operational task for the next 6–12 months.
Who wins, who loses
Upstage. First, proprietary search index — a strategic asset in global LLM competition. OpenAI relies on Bing (Microsoft), Anthropic on Brave Search; Perplexity has a partial in-house index. After this deal, Upstage is the only LLM company with a proprietary Korean / East Asian index. Second, distribution — Solar LLM usage scales to Daum's 20M MAU instantly. Third, valuation runway — next round plausibly opens at $1.5–2B.
Daum. A stagnant business under Kakao gets a new growth thesis under an LLM-first owner. AI Daum could rebuild engagement. Short-term cost in disentangling from Kakao infrastructure.
Kakao. Sheds an underperforming asset, gains LLM/search exposure through Upstage equity. KakaoTalk integration with Solar becomes the easy follow-on. Loses direct control of search/portal — an acceptable trade.
Solar LLM users. Korean-language quality improvement noticeable in 6–12 months. RAG accuracy in Korean retrieval likely +20–30%.
Naver. Direct impact limited by Daum's small share. But strategic risk: a real Korean LLM + search competitor exists for the first time. Naver's response: accelerate HyperCLOVA X upgrades, push Cue: monetization, deepen LINE Yahoo East Asia integration.
LG AI Research / Konan / Saltlux. Pressure to differentiate. LG can lean on EXAONE + LG U+ mobile data; Konan and Saltlux double down on enterprise RAG verticals.
Korean ad/media industry. Daum's ad business as a testbed for "AI advertising" — sponsored context inside LLM responses — is a first attempt in Korea. Publishers face a new licensing-renewal cycle including "AI training license" terms.
Past patterns — what worked, what didn't
Worked: Microsoft + Bing + OpenAI (2023–). Bing share didn't move much, but OpenAI users + Microsoft search infra became the Copilot growth engine. The Upstage + Daum analog. Difference: Microsoft only owns part of OpenAI; Upstage owns 100% of Daum.
Worked: Naver + LINE + Yahoo Japan (2019). Naver merged LINE into Z Holdings, building an East Asia super-app. Search + messenger + payments integration. A potential Upstage global-expansion playbook.
Failed: Yahoo + Tumblr (2013). Acquired for $1.1B; sold for $3M in 2019. Lesson: cultural and business-model mismatch destroys value. Upstage's "tech-first" culture vs. Daum's "media-first" culture is a real risk.
Failed: AOL + Time Warner (2000). The textbook bad merger. Upstage + Daum is much smaller and more focused, but the warning about tech-company-running-media-company applies.
Counter-plays
Naver. Likely accelerates HyperCLOVA X upgrades, considers a global English Cue:, deepens LINE Yahoo integration. Builds an "AI advertising" counter-model to whatever Daum + Upstage launches.
LG AI Research. EXAONE 4 acceleration. Index gap is the weakness — LG U+ mobile index + academic data as a partial fix. Builds an LG ThinQ + LG U+ super-app analog to Upstage + Kakao.
KoGPT (Kakao internal). Effectively wound down; Kakao gets LLM exposure via Upstage equity. KoGPT staff likely transferred to Upstage or redeployed.
OpenAI / Anthropic in Korea. OpenAI: ~10M ChatGPT MAU in Korea. Some Korean usage will migrate to Daum + Upstage. OpenAI is upgrading Korean fine-tunes; lack of native index is the gap. Anthropic's Korea entry is later — limited near-term impact.
Perplexity. Has Korean support but lacks a Korean-native index and LLM. Daum + Upstage is a direct counter inside Korea.
So what changes — by persona
Korean consumers. In 6–12 months, Daum's UI shifts from keyword search to AI answers. Similar to Perplexity, but more natural in Korean. Caveat: ad-injected context inside LLM answers requires user vigilance.
Korean creators / bloggers / publishers. Consent and licensing for content used in Solar training becomes the negotiation. Daum Cafe/blog operators likely get explicit opt-out. News publishers can demand a separate "AI training license" line in renewal contracts.
Korean advertisers. A potential new "AI advertising" category. Sponsored context embedded in LLM responses. Beta-only until disclosure rules under the AI Basic Act + Truth in Advertising Act are finalized.
Korean LLM startups. With Upstage now full-stack, differentiation must come from verticals — enterprise RAG, legal/medical/financial domain LLMs, voice/image multimodal niches.
Investors. Watch for Upstage's next-round valuation jump and Kakao's mark-to-market on its Upstage stake. Korea's LLM market consolidates into "Naver vs. Upstage + Daum" — capital-raise difficulty rises for the rest.
Regulators. KFTC clearance is the immediate watch. In the next 12 months, "AI advertising disclosure guideline" and "LLM training data copyright guideline" need clarity — likely a joint MSIT + MCST + KCC + KFTC effort.
Privacy / civil society. Transparency demands on how Daum user data (search history, mail, cafe activity) flows into Solar training. PIPA risk is the largest operational exposure. Civil-society oversight will sharpen over the next 12–24 months.
References
관련 기사
AI 트렌드를 앞서가세요
매일 아침, 엄선된 AI 뉴스를 받아보세요. 스팸 없음. 언제든 구독 취소.