Korean AI Services: HyperCLOVA X, Kakao, SKT, Exaone — 2026-09-26
Kakao confirmed the shutdown of its Kanana chatbot after MAUs collapsed 85%, refocusing AI efforts on KakaoTalk. Meanwhile Naver Cloud formally began building Korea's cybersecurity-focused sovereign AI models with 4,512 GPUs alongside LG, targeting frontier-level capability within 10 months. In the wider ecosystem, Seoul Economic Daily's reporting on stalled government-funded research and Wrtn's IPO push frame a tight ongoing debate over sustainable sovereign-AI funding.
Korean AI Services: HyperCLOVA X, Kakao, SKT, Exaone — 2026-09-26
Top developments
Kakao discontinues Kanana chatbot after 85% MAU collapse
Kakao has discontinued its generative AI chatbot Kanana roughly 1.5 years into development, without ever achieving a full public release. Chosun Ilbo reports Kanana's monthly active users had dropped 85% amid dominance from global rivals, and notes the setback adds to a pattern of Kakao AI struggles including earlier LLM failures. The company is now shifting its AI focus to integration within KakaoTalk, which matters as Kakao simultaneously prepares its "AI for All" national consortium service.
Naver Cloud kicks off cybersecurity-specialized AI model project with 4,512 GPUs
On September 22, a Naver Cloud-led consortium with 40 partners launched development of two cybersecurity-specialized AI foundation models for the government, deploying 4,512 GPUs over roughly 10 months (about 830TB of associated infrastructure planning). The models will be built in parallel on HyperCLOVA X and LG's EXAONE, each scaled toward 700 billion parameters, with plans to eventually export the security-specialized models overseas. Deputy Prime Minister and Science and ICT Minister Bae Kyung-hoon attended the kickoff meeting with Naver Cloud CEO Kim Yu-won.
Government-funded K-Fold bio AI model stalls: from 256 GPUs to two
Seoul Economic Daily reported on September 21 that K-Fold, a KAIST-developed bio AI model billed as a "Korean AlphaFold," is now running on just two GPUs after its 256-GPU government grant under the MSIT "AI-specialized foundation model" program expired — leaving retraining and upgrades effectively halted. The story spotlights the core sustainability problem of Korea's competitive, review-cycle-driven sovereign AI funding model, where research momentum is lost the moment a grant ends.
Sovereign AI foundation model teams shrink; Naver drops out
Local reporting this week noted that in the space of a year the MSIT "independent AI foundation model" project has changed its cohort three times, with only three teams remaining after Naver was not carried forward — the project narrative now covering Upstage, SK Telecom, NC AI, and LG AI Research among previously selected participants.
Local view
Korean-language coverage is sharply focused on the contrast between lavish infrastructure commitments and fragile research pipelines. Seoul Economic Daily's exclusive on K-Fold ("256 GPUs used, then 2 once support ended") set the agenda for local tech press. Naver Cloud framed its cybersecurity AI bet in sovereign-security terms, telling AI Times the goal is frontier-level capability beyond mere model size, and EBN covered the joint government–Naver consortium's 4,512-GPU deployment plan. Seoul Finance, in a 24th-anniversary special on Korea's AI ambitions, positioned the country as struggling to secure "AI G3" status after ChatGPT reshaped the field.
Context & numbers
- Kakao Kanana MAU fell 85% before its discontinuation
- Naver consortium: 4,512 GPUs, ~830TB, 40 partners, ~10-month timeline, two parallel models based on HyperCLOVA X and EXAONE, each targeting 700B parameters
- 2026 government AI budget: 10.1 trillion won, up roughly 57.8% year-on-year
- National AI infrastructure plans target 260,000 high-performance GPUs
On the radar
- Wrtn, valued above 1 trillion won as an AI unicorn, is formally pushing toward an IPO with plans to file for preliminary review next year following underwriter selection — expect more listing coverage.
- The 2026 National Assembly parliamentary audit is expected to scrutinize the homegrown AI foundation model project's market results, the 260,000-GPU operating plan, and child/teen chatbot safety measures.
- Kakao's post-Kanana strategy of embedding AI agents directly into KakaoTalk warrants watching, especially as its "AI for All" consortium beta approaches.
This content was collected, curated, and summarized entirely by AI — including how and what to gather. It may contain inaccuracies. Crew does not guarantee the accuracy of any information presented here. Always verify facts on your own before acting on them. Crew assumes no legal liability for any consequences arising from reliance on this content.
Powered by
