Korean AI Services: HyperCLOVA X, Kakao, SKT, Exaone — 2026-10-03
South Korea's homegrown AI foundation model programme enters a critical pivot phase, with the government weighing expansion into "frontier AI" development while three teams—SK Telecom, Upstage, and LG AI Research—remain in the race after over 200 billion won in cumulative investment. The "Everyone's AI" free public service rollout by SKT, Kakao, and KT faces real-world infrastructure constraints, with newly disclosed data showing 70% of government GPU allocations operating below 50% capacity.
Korean AI Services: HyperCLOVA X, Kakao, SKT, Exaone — 2026-10-03
Top developments
Government-backed model project faces capacity reality check
Over the past week, South Korea's Ministry of Science and ICT clarified that its sovereign AI foundation model (독파모, "independent foundation model") programme will continue through February 2027, rejecting speculation it would be halted. However, new data reveals systemic underutilization: according to ZDnet Korea (October 1), approximately 70% of government-allocated GPUs are operating at less than 50% capacity, exposing gaps in infrastructure management and utilization planning.
The second evaluation round in August confirmed three surviving teams: SK Telecom, Upstage, and LG AI Research. Over 200 billion Korean won has been invested in the project to date. The Ministry's 530 billion won target aims to achieve a homegrown model at 95% of global frontier performance.
Pivot to "frontier AI" signals scaling ambitions
Deputy Prime Minister Bae Kyung-hoon announced on September 30 that the government will pursue a parallel "frontier AI" development track alongside the foundation model programme, targeting frontier-class AI competing with global leaders. The Ministry is exploring a special-purpose company (SPC) structure to consolidate resources from multiple firms, with a third evaluation planned for early 2027.
This marks a strategic recalibration: rather than abandoning the foundation model initiative, Seoul will pursue a two-track approach—maintaining homegrown capability while attempting breakthrough-level performance through consolidated effort and industrial partnership. The Ministry has allocated 384.9 billion won toward securing "full-stack" AI technology, from foundational models to hardware accelerators.
AI agent race heats up: Naver, Kakao, SKT, KT vie for $251B market
On September 29, Chosun Ilbo reported that Naver, Kakao, SK Telecom, and KT are competing to deliver agentic AI services—systems that autonomously handle complex tasks like shopping recommendations and payments. The emerging market is estimated at $251 billion globally. This extends beyond the free "Everyone's AI" chatbot programme into higher-margin autonomous agent platforms.
National portals block foreign AI crawlers; data sovereignty tightens
On September 29, Naver and Daum—South Korea's two largest internet portals—announced they will block global AI companies from crawling their platforms to protect Korean-language data. This move reflects Seoul's broader sovereign AI strategy: preserving domestic content for local model training while limiting free training data access to foreign AI firms.
Domestic models achieve "notable" status in global epoch ranking
SK Telecom, Upstage, LG AI Research, and Motif Technologies—all participants in the second-stage homegrown AI evaluation—had their models recognized as "notable models" by Epoch AI's global ranking, placing South Korea third globally after the US and China in foundation model capability. According to Edaily (October 1), this validation demonstrates proof-of-concept for Korea's independent development path.
Local view
Edaily (Oct 1): Korean tech commentator Kim Hyun-a praised the achievement, writing "독파모로 AI 3위권 가능성 확인했다" (We've confirmed Korea's AI third-place possibility with homegrown models), signalling domestic media confidence that Korea's sovereign approach is viable at scale.
Seoul Economic Daily and AI Times emphasized the Ministry's commitment to both tracks—foundation models and frontier AI—as a "hedging strategy": continuing incremental capability building while pursuing breakthrough performance. Upstage's leadership responded to the restructuring proposal by stressing "기술혁신" (technological innovation) over raw model size.
Chosun Biz: Questioned the feasibility of the 530 billion won budget reaching frontier-class performance, noting that the gap between government ambitions and compute reality creates execution risk.
Context & numbers
- Government AI budget 2026: Approximately 10.1 trillion won (MSIT estimates 4.5–5.0 trillion won directly allocated to AI initiatives)
- Homegrown model programme cumulative spend: Over 200 billion won through second evaluation
- Frontier AI and full-stack technology (2026): 384.9 billion won allocation
- GPU underutilization: 70% of government-allocated GPUs operating below 50% capacity
- National AI Computing Center (Haenam): 15,000 GPUs planned; 2.5 trillion won total infrastructure investment; groundbreaking in August 2026
- Zeta (domestic AI chat app): 4.02 million monthly active users (February 2026); over 113.41 million hours monthly usage—more than double ChatGPT's time despite smaller user base
- ChatGPT (South Korea): 22.93 million monthly active users (February 2026), still dominating by MAU
On the radar
- Third foundation model evaluation: Scheduled for early 2027; outcome will determine which teams advance to SPC-based frontier AI development
- "Everyone's AI" beta and rollout: SKT, Kakao, KT consortiums targeting nationwide launch within the year (by end-2026); infrastructure readiness critical given GPU utilization constraints
- Samsung–Helix investment: $1 billion injection by Samsung Electronics and affiliates into Nvidia- and KKR-backed AI infrastructure firm (announced Sept 29), signalling Korean conglomerate pivot toward global AI infrastructure rather than purely domestic sovereign models
- Naver–Nvidia DSX expansion: Government-industry partnership to scale Naver's data center to 200 megawatts—expanding original 55 MW deployment—remains on construction timeline but faces GPU supply competition with global demand
Data freshness: All reporting and figures in this article are dated September 26–October 3, 2026. GPU utilization, foundation model evaluation outcomes, and government policy announcements are current as of October 3, 2026.
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
