S Korea removes Naver, NCSoft teams from sovereign AI model race

South Korea has removed teams led by Naver and NCSoft units from its national competition to develop a sovereign AI model, the Ministry of Science and ICT announced on January 15.

The three remaining teams are led by LG AI Research, SK Telecom, and Upstage.

The competition began with 15 applicants, narrowed to five finalists in August, and will continue with periodic evaluations and eliminations until two winners are chosen by 2027.

The program aims to build AI foundation models developed in Korea, amid concerns about reliance on foreign technology.

Questions have been raised by rival firms and some experts over whether several teams used non-Korean components, sparking debate about compliance with requirements to create open-source models independently.

The ministry said teams were evaluated on technical, policy, and ethical grounds.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

State resources committed to sovereign AI remain unclear despite 2027 deadline

So far, filings do not list a total budget for the sovereign model contest. Officials set KRW 1.9 trillion to secure 10,000 graphics processing units (GPUs) and KRW 213.6 billion for world-class large language model (LLM) projects 1.
The Ministry of Science and ICT proposed KRW 23.7 trillion for 2026, with KRW 4.46 trillion for AI transformation 2. Releases do not state compute credits or GPU pools for the teams led by LG AI Research, SK Telecom, and Upstage through 2027.
Available materials omit compute budgets, milestones, or scoring rubrics for periodic checks, so outsiders cannot tell if resources fit the plan to build foundation models without foreign technology by 2027.

Compliance audits gain traction as scrutiny over domestic origin intensifies

Naver plus NCSoft-led teams were removed. That has spurred demand for third-party checks covering open-source license compliance, data lineage, and component provenance.
Open-source software (OSS) auditing firms could tailor services to rules on domestic origin for independently built open-source models, with the Eclipse Foundation and its Open Regulatory Compliance Working Group offering guidance on regulations 3.
Korea plans KRW 87.1 billion for domestic AI semiconductor commercialization 4, so audit software providers can help fabless companies (chip designers that outsource manufacturing) plus downstream adopters (companies integrating these chips into products) clear audits and improve bids for public contracts that may require verified local supply chains.