Naver’s cloud booth at the first presentation of the “Sovereign AI Foundation Model” project, held on Dec. 30 at COEX in Gangnam, Seoul. (Yonhap)
SEOUL, Jan. 15 (Korea Bizwire) — Naver’s surprise elimination from the first round of South Korea’s government-backed independent artificial intelligence foundation model project has sharpened attention on how the internet giant will chart its AI strategy outside the state-led initiative.
According to the ICT industry on Thursday, Naver Cloud’s failure to advance in the initial evaluation stunned many observers, as Naver had long been regarded—alongside LG AI Research—as one of the two strongest contenders in the race to develop a homegrown foundation model.
Naver had unveiled ambitious plans to move beyond large language models toward an advanced multimodal system capable of processing speech and images. But its use of a Chinese-developed AI encoder—and, critically, the reuse of pre-trained model weights—proved decisive. Government officials ruled that this approach fell short of the project’s requirement for technical independence.
Ryu Je-myeong, second vice minister at the Ministry of Science and ICT, said evaluators expected teams to reset model weights and train systems anew using self-collected data, even when open models were referenced. “The issue arose from using existing weights as they were, which raised technical concerns,” he said.
Vice Prime Minister and Science Minister Bae Kyung-hoon echoed that stance in a social media post, urging eliminated teams to “accept the results with grace and return stronger.”
Although the government said it would later select one additional team in what amounts to a second-chance round, Naver indicated it would not reapply, saying it respected the government’s decision. Instead, the company is expected to pursue its AI ambitions independently.
Naver has already been accelerating overseas AI initiatives, including digital twin projects in Saudi Arabia, a sovereign large language model program in Thailand and the development of an AI data center in Morocco.
The decision has also sparked debate within the industry over how “from scratch” development should be defined. Critics argue that while the government emphasized scratch-built models as a core criterion, it failed to clearly delineate acceptable uses of open-source technologies, fueling confusion around Naver’s disqualification.
Ryu said the ministry would refine the definition of “from scratch” in the next evaluation phase after consulting academic and industry experts, to prevent similar disputes going forward.
Kevin Lee (kevinlee@koreabizwire.com)
