A critical assessment of the $72M Series C that made WRTN Korea’s first AI-service “unicorn” (₩1T+) — announced the same week its first regulatory filing revealed a 2025 operating loss (₩58.8B) larger than its entire revenue (₩47.1B), on a business that resells GPT, Claude and Gemini it does not own.
WRTN owns no model — it resells GPT/Claude/Gemini inference and gives its portal away free. With a ₩58.8B loss on ₩47.1B revenue and zero disclosed gross margin, there is no public evidence it makes money on the AI it sells. The thesis rests on a promised profitability “inflection” the filings actively contradict.
Scatter Lab’s Zeta has 1.4M MAU (2.5× Crack), longer engagement, and ₩2.6B operating profit — yet raised only ~₩50B. WRTN’s premium is built on forward guidance (>₩200B for 2026) front-loaded by a three-month-old US launch, not on category leadership or profit.
~10% of Crack users are minors, in-app payments have no total cap, adult-adjacent roleplay runs on suppliers who ban that content, and Korean and US regulators are drafting AI-companion minor-safety rules. WRTN’s monetization engine sits at the intersection of every one of those pressure points.
Key Finding: WRTN just became Korea’s first AI-service unicorn — but the same week’s regulatory filing shows a company whose 2025 operating loss (₩58.8B) was larger than its entire revenue (₩47.1B), monetizing rented frontier models it does not own, in a category where its closest rival (Zeta) is both bigger and already profitable. A ₩1T valuation is being paid for a thin reseller layer whose margins are set by its suppliers and whose growth engine — minors-heavy, uncapped-payment AI roleplay — is a regulatory bullseye.
WRTN gives frontier AI away free on its portal and resells inference through Crack/OOC. Every message a user sends is a variable cost paid to a supplier who is also a potential competitor. Here is the value chain — and where WRTN actually sits in it.
OpenAI, Anthropic, Google own the models. They set inference prices, rate limits and content policies.
A UX skin — portal + “playable anime” roleplay — on top of rented inference. Pays COGS per message.
A small heavy-spending cohort funds most revenue via subscriptions + uncapped in-app “superchat.”
Suppliers control input cost and can compete downstream; whales churn. Margin is set by parties WRTN doesn’t control.
WRTN publishes no gross-margin figure, no proprietary-model roadmap, and no evidence it can defend margin if frontier-model API prices, rate limits, or content policies move against it. A consumer AI aggregator with no model of its own is a thin reseller layer whose unit economics are set by suppliers it does not control — and whose most lucrative content those same suppliers explicitly restrict.
WRTN’s first statutory filing shows 2025 revenue of ₩47.1B against an operating loss of ₩58.8B — roughly ₩1.25 spent for every ₩1 of revenue, with the loss doubling in the same year revenue grew 15×. The “inflection point” toward profitability is asserted by management, not demonstrated by the numbers. A >₩1T valuation on ₩47.1B trailing revenue is ~21× trailing sales for a deeply loss-making token reseller.
AI character chat — “playable anime” roleplay. The actual revenue engine; overseas sales now exceed domestic.
OOC’s US launch spike. No retention, DAU/MAU or ARPU disclosed — the numbers that separate a business from a moment.
Crack’s single-payment cap is ₩126,000 (~$91) with no lifetime limit — whale monetization by design.
>₩1T valuation on ₩47.1B revenue, priced off unaudited 4× forward guidance.
A national celebrity ad push drove brand lift — real distribution strength, but also real customer-acquisition cost.
Adult-adjacent roleplay runs on models whose usage policies restrict exactly that content. A policy change could kneecap the product.
WRTN’s closest rival is bigger and already profitable; the global incumbents got there first; and the “magic” is rented frontier models anyone can rent.
Korea’s #1 AI character-chat app: 1.4M MAU (2.5× Crack), the longest-used AI app in Korea, and profitable — ~₩22.9B revenue and ₩2.6B operating profit — on only ~₩50B raised. WRTN’s closest rival is bigger in the monetized category and makes money, while WRTN loses ₩58.8B.
Character.AI defined the category and was effectively absorbed into Google (~$2.7B deal, 2024); Replika is the incumbent AI-companion abroad. OOC competes directly with both for the North American anime/roleplay audience — as a late entrant.
Korean AI search, ~₩93B raised cumulatively, 14M users across 220+ countries — competing for the “AI portal/search” positioning WRTN uses as its free top-of-funnel.
Naver Webtoon is entering AI character chat (2026) armed with IP and distribution; Kakao and Naver loom with reach WRTN cannot match. Platform-owner threats in WRTN’s home market.
The read-through: WRTN is domestic #2 behind a bigger, profitable Zeta; late abroad against Character.AI and Replika; and its differentiation — “playable anime” storytelling — is a UX skin over rented models that rivals can copy. The category tailwind (Antler’s “loneliness epidemic → AI entertainment” thesis) is real; a defensible WRTN-specific edge is not yet visible.
Seven structural risks that a $72M round and a unicorn headline do not resolve.
WRTN wraps GPT/Claude/Gemini with no proprietary model. Inference is variable COGS paid to suppliers who set prices and can compete downstream. No gross-margin disclosure; a ₩58.8B loss suggests thin or negative contribution margins today.
Operating loss (₩58.8B) exceeded revenue (₩47.1B) and doubled year-over-year. ₩100B raised buys limited runway if losses scale with usage. The “inflection point” is a claim, not evidence.
Crack was named in Korea’s Oct 2025 National Assembly audit for youth-protection gaps; ~10% of users are under 18; single payments cap at ₩126,000 with no total limit. Korea and the US are moving toward AI-companion minor-safety laws. An incident is a franchise risk, not a fine.
Domestic #2 behind a bigger, profitable Zeta; late abroad vs. Character.AI/Replika; and the underlying capability is rented frontier models anyone can rent. “Playable anime” is a UX skin, easily copied.
Monetization leans on uncapped in-app spending by a small heavy-user cohort in a high-churn category. OOC’s “₩10B/month in 3 months” is a launch spike with no reported retention or ARPU behind it.
>₩1T (~21× trailing sales) on unaudited 4× forward guidance and a three-month-old US launch. A guidance miss or US-growth stall re-rates hard; Bloomberg’s outlier ~$870M figure hints at fuzzy valuation reporting.
Content-heavy roleplay — some adult-adjacent — runs on OpenAI/Anthropic/Google models whose usage policies restrict exactly this content. A supplier policy change could kneecap the product overnight; self-hosting an open model would cut quality or raise cost.
WRTN has executed impressively on distribution — free portal to viral character chat to a fast US launch, with real (if launch-spiked) revenue and a brand most Korean AI startups can’t match. But strip the unicorn headline and you find a deeply unprofitable aggregator with no proprietary model, negative-looking unit economics it won’t disclose, a bigger profitable competitor at home, and its monetization concentrated in exactly the content category regulators in Korea and the US are moving to restrict. The Series C buys runway and a valuation trophy; it does not resolve whether reselling other people’s AI is a durable, defensible, compliant business. Watch the 2026 audited margin — not the top-line guidance — for whether the “inflection point” is real.
Based on publicly available information around the August 26, 2026 announcement. ₩-denominated figures (₩100B round, >₩1T valuation, ₩230B cumulative, ₩47.1B revenue / ₩58.8B loss) are the reliable spine; USD figures diverge across outlets and Bloomberg’s ~$870M valuation is an uncorroborated outlier. Company guidance and MAU claims are labelled as such.
