In 1997, years before Naver existed, a small Seoul company handed Koreans something nobody else would give away: a free email address. That company was Daum. Within a few years, the Daum portal had become the front page of the Korean internet — the place where people checked mail, joined fan clubs, argued about politics, and read the world’s first vertical-scrolling comics.
Today it holds 2.94 percent of Korea’s search market.
Yet in January 2026, an artificial intelligence startup agreed to buy it anyway, in a deal reportedly valued at roughly 200 billion won. The buyer did not want the traffic. Instead, it wanted thirty years of Korean sentences.
This is the story of how Korea’s first portal lost almost everything, and why the wreckage turned out to be one of the most valuable assets in the country’s AI race.
Daum Communications was founded on February 16, 1995, by Lee Jae-woong and two partners, with 50 million won in capital. Curiously, it started as an art site. However, the pivot came fast.
In 1997, the company launched Hanmail — free web-based email. To understand why that mattered, consider what Korean internet users had before. Most of them paid for dial-up services such as Chollian, Hitel, and Nownuri, where every minute cost money. Hanmail was free, permanent, and accessible from anywhere.
Koreans signed up in the millions. As a result, the company renamed itself Daum Communications in July 1999 and listed on the KOSDAQ that November at 11,100 won per share. At the height of the dot-com bubble, the stock touched 540,000 won.
Then came the feature that turned an email service into a culture. Daum Cafe let anyone spin up a private community in seconds — no coding, no server, no cost. For foreign readers, the closest analogy is Facebook Groups, except Daum Cafe arrived a decade earlier and went far deeper. Korean fan clubs, hiking groups, university alumni networks, apartment complexes, and early K-pop fandoms all organized inside cafes. Many still do.
Meanwhile, Daum built Agora, an open debate board that became Korea’s de facto public square. In 2008, when protests over American beef imports filled central Seoul with candlelight, much of the organizing and argument ran through Agora first. Academics still study it as a landmark case of internet activism in South Korea.
The company even relocated its headquarters to Jeju Island, an unusual move for a Korean tech firm. Planning began in 2004, the Daum Global Media Center opened in 2006, and the full move to the Space.1 building followed in 2012. For a while, the Daum portal looked like a company writing its own rules.
Through 2004, it was the most-visited portal in Korea.
Foreign readers usually underestimate this product, so it deserves a proper explanation.
A Korean “cafe” is not a chat room. Rather, it is a miniature private website with membership tiers, permission levels, posting requirements, and internal reputation systems. Joining a serious cafe often means applying. Some require an introduction post. Others demand a minimum number of comments before you can read the good boards.
Because of that structure, cafes became repositories of extremely specific expertise. Want honest reviews of a particular apartment complex in Suwon? There is a cafe. Looking for used camera lenses, secondhand strollers, or advice on a rare medical condition? Cafes again.
Crucially, this content is largely invisible to Google. It sits behind membership walls, in Korean, on a platform Western crawlers barely index. Consequently, an enormous slice of Korea’s practical knowledge exists nowhere else.
Tistory added a second layer. Launched in 2006, it became the blogging platform for Koreans who wanted control over their own design and domain. Developers, photographers, and hobbyists built long-form archives there. Brunch, added later under Kakao, attracted essayists and aspiring authors instead.
Notice the pattern. Each of these products failed commercially by portal standards. Nevertheless, each one produced exactly the kind of text that language models are starving for: unpaid, unedited, deeply human Korean writing about real life.
Nobody was thinking about training data in 1999. Even so, that is what got built.
In 2003, Daum launched a comics section called Manhwa Sokui Sesang. Rather than reproducing print pages, it published comics designed to be read by scrolling vertically on a screen. That single format decision created the webtoon.
Early Daum artists became canonical. Kang Full, Yoon Tae-ho, and Chon Kye-young all built their reputations there. Yoon’s Misaeng later became one of Korea’s most influential dramas. In other words, the Daum portal seeded an industry that now generates billions of dollars, as covered in our breakdown of the Korean webtoon industry in 2026.
Nevertheless, Daum never captured the value. Kakao launched KakaoPage in 2013 with far heavier investment, and after the two companies merged, resources flowed there instead. Daum Webtoon was eventually rebranded as Kakao Webtoon. Finally, in March 2026, the Kakao Webtoon link disappeared from Daum’s homepage altogether.
The platform that invented the format no longer links to it.
Something similar happened with web fiction, another category Korean portals pioneered before the global market caught up — a shift we traced in our report on Korean web novels going global.
Daum did not die in a crash. Instead, it bled out over twenty years, and the causes are unusually instructive.
Naver solved a problem Daum ignored. In 2002, Naver launched Knowledge iN, a user-generated question-and-answer database. At the time, the Korean-language web was thin, so search engines returned very little. Knowledge iN manufactured the content that search needed. Consequently, Naver’s results simply became more useful, and users migrated.
Mobile arrived, and Daum was slow. Smartphones reshuffled Korean internet habits between 2010 and 2014. Meanwhile, KakaoTalk captured messaging almost overnight. Daum’s mobile products were competent but rarely essential.
Policy decisions cost it traffic. Daum abolished real-time search rankings in 2020, then narrowed news search results to registered content partners in 2023. Both moves were defensible on principle. Still, each one removed reasons to visit.
The numbers tell the rest. Naver held 76.69 percent of Korean search by 2014. According to InternetTrend data reported by The Korea Times, the 2025 breakdown looked like this: Naver 62.86 percent, Google 29.55 percent, Microsoft Bing 3.12 percent, and Daum 2.94 percent.
Korea’s first portal now ranks fourth in its own country, behind Bing. At its peak, by contrast, it commanded somewhere between 30 and 40 percent.
For the fuller picture of how the winner defended that ground, see our analysis of how Naver held off Google in the AI search war.
International readers may recognize the shape of this story. Yahoo and AOL followed similar arcs, after all. However, there is one important difference. American portals lost to a better search engine from the same open web. The Daum portal, by contrast, lost inside a closed ecosystem where the winner owned the content as well as the index. Naver did not merely rank Korean information. Instead, it manufactured and hosted it.
That distinction matters now, because it explains why Korea’s second-place archive stayed intact rather than dispersing across the open internet.
On October 1, 2014, Daum Communications merged with Kakao to create Daum Kakao. On paper, it looked like a rescue. In practice, it was closer to an absorption.
The structure tells you everything. Kakao was privately held and wanted a public listing, whereas Daum was already listed on the KOSDAQ. Therefore the merger handed Kakao a shortcut to the public markets. Less than a year later, in September 2015, the combined company dropped “Daum” from its name entirely.
Afterward, Kakao expanded relentlessly. The group eventually operated more than 140 affiliates spanning banking, taxis, payments, games, music, and entertainment. Daum, however, sat inside that empire as a legacy asset — maintained, occasionally redesigned, and never central.
Foreign founders often misread this pattern as failure. Yet Korean conglomerates routinely acquire platforms for their licenses, listings, or user bases rather than their products, a dynamic worth understanding before raising money here, as our guide to Korea’s startup ecosystem explains.
By 2025, Kakao had a new problem. Chief Executive Chung Shin-a took over with a mandate to simplify, and the affiliate count fell from roughly 140 to about 90 through closures and sales.
Daum was on the list.
In May 2025, Kakao carved the portal out of its content division into a separate subsidiary named AXZ. Operational separation finished on December 1, 2025, ending eleven years inside Kakao. Reporting from The Korea Times described the new company as an attempt to rebuild Daum as a content platform rather than a search engine.
The independent Daum portal moved quickly. It shipped an AI chatbot called DD, relaunched a feed product as Loop, added short-form drama, and introduced TimeTalk, a comment system that expires after 48 hours with AI moderation. In March 2026, it even revived real-time trending searches after a six-year absence, this time with AI filtering.
Above all, the strategy was clear: stop fighting Naver on search, and rebuild around communities the company still owns.
Because those communities never actually left. Daum Cafe, Tistory blogs, and Brunch essays remained active throughout the decline, quietly accumulating text.
On January 29, 2026, the boards of Kakao and Upstage approved a memorandum of understanding. Upstage would take full ownership of AXZ through a share swap, while Kakao would receive a minority stake in Upstage. The Korea Herald put the value at roughly 200 billion won, or about 140 million dollars.
Upstage is not a portal company. It builds Solar, a Korean large language model, and it became Korea’s first generative AI unicorn after a Series C that valued it around 740 billion won. Subsequently it was selected for the government’s sovereign AI program, backed by roughly 400 million dollars in national funding — a decision we examined in detail in Korea’s sovereign AI bet on Upstage. You can see the company’s model lineup on its own Upstage newsroom.
So why buy a portal with a 2.94 percent share?
The answer is data. Korean is a mid-resource language for AI training, and the good material is locked away. Naver will not license its ecosystem to a competitor. KakaoTalk conversations are private. Government open-data sets exist, but they are limited and heavily curated. Consequently, anyone building a genuinely Korean model faces a shortage of natural, messy, conversational text.
Daum has exactly that, and it has thirty years of it. Cafe threads capture how Koreans argue. Tistory posts capture how they explain things. Brunch essays capture how they write when they are trying to write well. News archives supply the formal register. Taken together, it is a corpus that cannot be recreated at any price, because the communities that produced it no longer behave the same way.
Upstage Chief Executive Kim Sung-hoon framed the logic around distribution rather than scraping, saying the combination would create an environment where more users could experience AI naturally.
Both readings are probably true. The portal supplies training material and a consumer surface at the same time.
Pricing a corpus is genuinely hard, so it helps to look at the comparables.
Reddit signed data licensing agreements with Google and OpenAI reportedly worth about 60 million dollars per year each. Shutterstock, meanwhile, has licensed its image library to multiple model developers. News publishers from Axel Springer to the Financial Times have struck similar arrangements. In every case, the buyer paid for access rather than ownership.
Upstage did something different. It bought the whole company, which means it acquired the corpus, the pipeline that keeps generating new text, and roughly ten million weekly users as a distribution channel. For about 140 million dollars, that is arguably cheap.
Yet the risks are real, and investors should weigh them honestly.
Ownership does not automatically settle consent. Furthermore, Korea’s Personal Information Protection Commission has been increasingly active, and user backlash could force opt-outs that thin the dataset. Model quality is not guaranteed either. Solar must compete with global systems that keep improving at Korean without any Korean data partnership at all. Finally, Upstage plans an initial public offering with valuation targets between 2 and 4 trillion won, which leaves little room for execution failure.
Still, the strategic logic is difficult to argue with. Compute can be rented, and engineers can be hired. Thirty years of Korean community writing, on the other hand, can only be bought once.
The rebuild started immediately. In July 2026, Daum launched an AI Summary beta powered by Upstage technology, aimed at the roughly ten million weekly users the portal still reaches.
Later that month, the corporate identity followed the product. Shareholders approved renaming AXZ to Daum Inc. on July 20, and the change took effect on July 27. Chief Executive Lee Gun-soo said the company would inherit more than thirty years of accumulated trust while combining it with AI technology.
That framing is careful, and it needs to be. Not everyone is comfortable.
Tistory bloggers and Brunch writers have raised pointed questions about whether their work now trains a commercial model, and under what consent. Korea’s platform terms generally grant broad service licenses, so the legal position is likely defensible. Nevertheless, the ethical debate is live, and it mirrors arguments happening in publishing markets worldwide.
Meanwhile, the company that once owned Daum is going through its own dismemberment. On August 21, 2026, Kakao announced a split into two entities: KakaoAI, holding KakaoTalk and consumer AI services, and KakaoX, holding payments, mobility, and entertainment. Korea JoongAng Daily reported a 36-to-64 share split, with completion targeted for January 1, 2027. Investors were unimpressed, and the stock fell 12 percent on the announcement.
In short, the empire that absorbed Daum in 2014 is now taking itself apart along the same lines.
Practically speaking, yes, though selectively.
Daum Cafe remains genuinely useful. For niche, local, Korean-language information, it often beats both Google and Naver. Apartment complexes, expat parenting groups, hiking clubs, and secondhand markets all live there. A translation extension handles most of the friction.
Tistory is worth knowing if you publish in Korean. Unlike Naver Blog, Tistory allows custom domains and full HTML control. Therefore it indexes reasonably well on Google, which matters if you want reach beyond the Korean walled gardens.
Daum Mail is legacy. Korean institutions still occasionally recognize hanmail.net addresses, but there is no reason to open a new account today.
Daum search is optional. Its AI summary beta is competent for Korean-language queries. Nevertheless, Naver still returns better local results, and Google still wins for anything international.
One caution belongs here. Since the Upstage acquisition, anything you post to a Daum property may plausibly become model training material. Read the current terms before you publish something you would not want in a corpus. That advice applies to Naver and Kakao too, of course, but the Daum portal has now made the intent explicit.
Three lessons stand out, and none of them are only about Korea.
First, distribution decays faster than data. Daum lost its audience over two decades. Its archive, though, kept compounding the entire time. In an AI market, the archive turned out to be the durable asset, while the traffic was the perishable one.
Second, Korea’s internet produces single winners. Naver takes search, Kakao takes messaging, Coupang takes commerce. Second place is not a stable position here, as we have seen repeatedly across sectors including live commerce. Foreign investors who model Korea like the United States, with room for a strong number two, tend to be disappointed.
Third, watch what gets sold, not what gets announced. Kakao’s press releases in 2026 emphasized AI ambition. Its divestitures, however, revealed the real strategy: shed everything that does not defend KakaoTalk.
There is one final irony worth sitting with. Daum spent twenty years being told it had lost the internet. Then the technology everyone assumed would finish it off — generative AI — turned out to need precisely what Daum had been accumulating while it lost.
Whether Upstage can convert that corpus into a competitive Korean model remains genuinely unproven. GPU capacity is scarce, Naver is entrenched, and global models keep improving at Korean. Still, the bet is a coherent one, and it reframes what a declining platform is actually worth.
Korea’s first portal is no longer trying to win the internet. Instead, it is trying to become the thing that teaches a machine to speak Korean.
That may be a smaller ambition. Even so, it is a far more interesting one.
Land at Incheon, open your phone, and the very first thing that breaks is the…
In January 2026, Korea's government published a shortlist that did not include Naver. Three organisations…
At 7:30 on a summer evening in July 2026, a fishing crew off Bangeojin hauled…
For centuries, the Korean fire festival season opened the same way. On the eve of…
The bowl arrives still boiling, and nobody at the table asks where the bird came…
Nobody comes to South Korea expecting a desert. Yet roughly 19 kilometres from the North…