South Korea has moved to tighten the definition of “sovereign” artificial intelligence as it accelerates a state-backed effort to build a fully homegrown large language model—an ambitious bid to narrow its gap with the U.S. and China in the global AI race.
On Jan. 15, the Ministry of Science and ICT said it had selected LG AI Research, SK Telecom and startup Upstage as finalists in the first round of evaluations for its Sovereign AI Foundation Model Project, a multiyear initiative designed to establish a domestically developed foundation model.
Notably absent from the list was Naver Cloud, the cloud-computing arm of South Korea’s dominant internet company. Despite strong benchmark performance, Naver was excluded on qualitative grounds tied to concerns over technological independence. NC AI, the artificial intelligence unit of game publisher NCSoft, was also eliminated for failing to meet minimum scoring thresholds.
According to the ministry, Naver’s model relied in part on architectural components and pretrained weights derived from Alibaba’s Qwen model. That dependence, officials said, ran counter to the government’s strict criteria for sovereignty, which require participating systems to be developed entirely in-house—from initial architecture design through large-scale pretraining.
“While Naver Cloud demonstrated competitive technical performance, expert reviewers raised serious concerns about the limits of its technological autonomy,” the ministry said in a statement. “After comprehensive review, the committee concluded that the model did not meet the core requirements of a domestically developed foundation model.”
The evaluation process assessed candidates across three dimensions: benchmark performance, expert review and user experience. LG AI Research ranked first overall, posting the highest scores in all categories, while SK Telecom and Upstage also cleared the required thresholds.
The government had initially planned to eliminate only one consortium. Naver’s disqualification on sovereignty grounds resulted in two teams being removed, prompting officials to announce plans for a wildcard selection later this year. One additional “elite team” will be chosen from previously eliminated applicants and other consortia to join the program.
Companies selected for the project gain access to government-funded GPU infrastructure, high-quality training data and the right to use the official “K-AI Enterprise” designation—an endorsement intended to strengthen their position against global AI leaders such as OpenAI and Google.
Policy analysts say the move underscores South Korea’s growing urgency to assert control over core AI technologies, even at the cost of sidelining national champions. While the country has long been a leader in semiconductors and consumer electronics, it has lagged behind the U.S. and China in large-scale AI model development.
By drawing a hard line on technological independence, Seoul appears to be signaling that its late entry into the AI arms race will be defined less by speed than by sovereignty—a high-stakes bet that domestic control, rather than rapid adaptation of foreign models, will ultimately determine its place in the emerging AI order.
