spoonai
TOPUpstageDaumKakao

Upstage Finalizes Daum Acquisition — A Korean LLM Company Just Took a Top-Tier Portal

On May 7, Korean LLM startup Upstage finalized the acquisition of AXZ, the Kakao subsidiary that runs the Daum portal. Stock-swap deal, four months of due diligence after a January MOU. Plan: fuse the Solar LLM with Daum's search and content corpus into a 'context-aware AI portal' to challenge Naver and Google.

·9분 소요·Electronic Times (etnews.com)Electronic Times (etnews.com)
공유
Upstage acquires Daum portal — Solar LLM-powered Korean AI search
Source: Asia Today

Korea's AI portal war just kicked off — Upstage took Daum

Here's the deal: on May 7, Upstage finalized the acquisition of AXZ, the Kakao subsidiary that runs the Daum portal. After a January MOU and four months of due diligence, the binding agreement was signed; the only remaining step is KFTC (antitrust) clearance. This isn't a routine IT acquisition. It's the first time a Korean LLM startup has acquired a major-tier portal, and Upstage has laid out a blueprint to fuse its Solar LLM with Daum's search engine and content corpus into what it's calling a "context-aware AI portal." Korea's search market — split today between Naver and Google — is about to get a third axis: Korean-built AI search.

The cast — Upstage, Daum, Kakao, Naver, Google

Upstage first. Founded in 2020 by CEO Sung Kim (former NAVER Clova lead, NYU professor), Upstage is a Korean-soil LLM lab. They shipped their own model "Solar" in 2024 — Korean-tuned 32B and 70B variants. December 2025's Solar 2.0 (150B parameters) topped Korean MMLU above GPT-5.3. Cumulative funding: ~500B KRW. As of April 2026 the valuation sits near 2.5T KRW. With this Daum deal Upstage transforms from a B2B SaaS LLM company into a hybrid "consumer portal + LLM" operator.

Daum is a 1995-vintage Korean web portal. It used to lead Korea's market until Naver overtook it; share fell from 30% → 5%, recovering to ~4% by May 2026. Kakao acquired Daum in 2014 but the synergy never materialized; from 2024 the company has been weighing a sale or spin-off. Today: ~25M DAU, ~4% search share, plus a deep content base across mail, news, cafes, and communities.

Kakao is the seller. At the 2014 merger Kakao's market cap was ~5T KRW; eleven years later it's ~8T KRW — roughly +60%, well below sector. The "weak synergy with KakaoTalk / Pay / Mobility" critique never went away. Carving out Daum lets Kakao refocus on its core. Importantly, the deal is structured as a stock swap: Kakao receives ~25% of Upstage rather than cash.

Naver dominates Korean search at ~60% share. On AI search it has been pushing CUE: chatbot integration. The Upstage-Daum deal is the most direct threat to its position — pointing toward an "AI search duopoly" instead of a Naver monopoly. Naver was -3% on May 8, reflecting that concern.

Google holds ~30% of Korean search. Korean-language NLP has been a structural weakness vs Naver, and Solar's win on Korean MMLU adds another headwind. Google's edge remains global index depth.

The hard numbers — stock swap, Solar + Daum, context-aware portal

Deal structure and integration plan in one table.

Item Detail Implication
Deal structure Stock swap No cash; Kakao gets ~25% of Upstage
MOU Jan 2026 Four-month diligence
Binding agreement May 7, 2026 Acquisition finalized
Remaining step KFTC merger review Expected 3-6 months
Daum assets Search, mail, news, cafes, content DAU ~25M
Solar integration target Q3 2026 Search and news first
Headcount Daum ~1,200 + Upstage ~500 ~1,700 combined
Combined post-deal valuation ~5T KRW (est.) 2.5T pre + synergy premium

The headline of the structure: stock swap means Kakao is now an Upstage shareholder. By taking equity instead of cash, Kakao is signaling that "Kakao should also have a seat in the AI portal future." Kakao's own LLM (KoGPT) lags Solar on Korean — the indirect-investment route gets Kakao back into the game.

The "context-aware portal" concept matters. Sung Kim framed it as a three-stage evolution: keyword search → intent search → context search. Keyword: type "best restaurants" → get a list. Intent: ChatGPT-style → "good-vibes restaurant near Gangnam tonight" → get a recommendation. Context: "book Mom's-day dinner reservation for OO" → factor in family composition, last year's restaurant choice, budget, time-of-day. Training Solar on Daum mail / cafe / blog data is what makes context search possible — that's the bet.

The first-order problem is privacy. Training an LLM on Daum users' mail / cafe / blog content requires consent flows. Korea's Personal Information Protection Act plus the AI Basic Act (effective 2026) impose strong restrictions on training with sensitive personal data. How Upstage handles consent ("opt-in" vs "opt-out") is the integration's pivotal variable.

What each side gets — Upstage, Kakao, Daum users, the Korean government

Upstage gets three things. First, distribution. A B2B SaaS company suddenly owns a 25M-DAU consumer channel — direct user-side exposure for Solar. Second, training data: Daum cafe / blog / news comments add up to ~500TB of Korean-language corpus, a real lever to push Solar's Korean performance further. Third, an IPO option. With combined valuation reaching 5T KRW, a 2027-2028 listing in Seoul or NY becomes plausible.

Kakao gets two. First, non-core asset cleanup, refocusing on KakaoTalk / Pay / Mobility. Re-rating from a P/E of ~16x toward 22x is on the table. Second, indirect AI re-entry: 25% of Upstage gives Kakao access to the Korean AI portal market without needing to win at LLM development on its own.

Daum users get "AI search + 25 years of accumulated content." Cafes, blogs, and mail stay in place and now feed AI search. The open question is consent design: opt-in protects user privacy at the cost of slower data acquisition; opt-out is faster but invites backlash.

The Korean government gets an "AI sovereignty" win. A Korean LLM operating Korea's #2 portal cuts dependence on ChatGPT / Gemini for sovereign data flows. The April-passed AI Basic Act (effective 2026) explicitly favors Korean AI infrastructure — this deal aligns with policy.

Historical comps — Kakao-Daum (2014) and Naver-LINE

Comp #1: Kakao acquires Daum, 2014. The thesis: KakaoTalk (messenger) + Daum (portal) makes Korea's #1 internet company. Result, both ways: Kakao's market cap 3.6x'd post-deal (5T → 18T KRW), but Daum's own share kept falling from 30% to 5%. The "messenger + portal synergy" never showed up. Whether Upstage's "LLM + portal synergy" rhymes with that failure is the central question.

Comp #2: Naver spins out LINE, 2014. Naver carved LINE out into a Japanese subsidiary that listed on TSE. Result: success — LINE locked Japan's messenger #1, and Naver Group market cap went from 6T → 30T (5x). The "carve out, go global" template. If Upstage takes Solar+Daum into Southeast Asia post-integration, that's the winning analog.

Comp #3 (positive): Microsoft acquires GitHub, 2018 ($7.5B). MS bought GitHub for developer distribution and synergy with Azure. The Upstage analog ("Solar + Daum = Korea's MS Azure + GitHub") is a stretch — Daum's reach is ~1/100 of GitHub's — but the structural logic is similar.

Bust comp: SK divests 11Street, 2017. SK gave up the "e-commerce + telco synergy" thesis; both sides weakened afterward. If Kakao's Daum sale follows that pattern, both Kakao and Upstage face headwinds. If it follows the "non-core cleanup" pattern, both win. That's the next 18 months.

One more reference: Sundar Pichai's Google + Bard integration, 2024. Putting Bard (now Gemini) at the top of search created AI Overviews. Result, both ways: search clicks softened (users stopped at the Overview), but ad CPMs rose. Solar+Daum likely follows the same arc.

Counter-plays — Naver CUE, Google Korea, Kakao KoGPT, cloud big three

Naver CUE: is the closest competitor — already in production since 2024. Strong on Korean NLP, but model size is smaller (45B vs Solar 2.0's 150B) and training-data diversity lags. Naver replied within 24 hours: a "CUE: Pro" announcement on May 8. The Korean AI search duopoly is officially live.

Google holds 30% in Korea. Gemini 2.5 Pro Korean integration is the defense. Korean-language NLP and Korean-content indexing remain structural disadvantages — expect more localized investment (Seoul AI center, Korean-specific fine-tunes).

Kakao KoGPT effectively merges into the Solar story via the 25% stake. The standalone KoGPT release scheduled for May 8 was deferred. Kakao is shifting from "build our own LLM" to "invest in and use a partner LLM."

Korean cloud big three (Naver Cloud, KT Cloud, NHN Cloud) get mixed effects. Negative: Solar increasingly runs on Naver Cloud, taking LLM-hosting revenue from rivals. Positive: a brand-new "AI portal" category lifts the overall Korean cloud TAM, so all three benefit at the macro level.

International competitors — OpenAI ChatGPT, Anthropic Claude, Google Gemini — are stuck at 5-10% combined Korean share due to the Korean-language quality gap. Solar+Daum makes that gap wider, raising barriers for English-first models.

So what changes — Korean users, developers, content creators

For Korean users, two shifts. First, the Daum search interface will likely move toward "AI chat first," with Solar answers above keyword results. If Naver search reads as "AI Overviews + ads + blogs," Daum will read as "AI chat + user context + content." Second, Daum mail / cafe / blog could become AI training data. The opt-in / opt-out call decides how big the user-backlash risk gets.

For Korean developers, "Solar API + Daum content API" becomes a new building block. Korean LLM call cost likely drops to ~1/3 of GPT-5.5 Instant — a real boost for Korean startups in e-commerce, education, travel. Watch for the Solar-lock-in cost: migrating to OpenAI / Claude when going global isn't trivial.

For Korean content creators (bloggers, YouTubers, media), it cuts both ways. Negative: AI answers above the fold means lower click-through to source content. Positive: Upstage may license training data from creators — a new revenue stream worth watching.

For Korean e-commerce, finance, and media, Daum's value as an ad medium changes. Higher user trust from AI integration could lift CPMs. But the new question is "search-result vs AI-answer" ad slot allocation — fresh negotiation territory.

The biggest strategic takeaway is global. The "Korean AI portal" model is now an exportable template for non-English markets — Thailand, Vietnam, Indonesia. If Upstage replicates the Daum playbook with portal acquisitions in Japan or Southeast Asia, an entirely new "non-English AI search without OpenAI dependency" category emerges. That's the 24-month strategic fork worth tracking.

References

관련 기사

무료 뉴스레터

AI 트렌드를 앞서가세요

매일 아침, 엄선된 AI 뉴스를 받아보세요. 스팸 없음. 언제든 구독 취소.

매일 30개+ 소스 분석 · 한국어/영어 이중 언어광고 없음 · 1-클릭 해지