Companies participating in Korea's sovereign AI foundation model project are accelerating real-world industrial deployment ahead of a second evaluation. As the government has presented not only model performance but also real-world applicability as a key criterion in the sovereign AI model evaluation, the firms are moving to demonstrate their validation capabilities across various fields such as manufacturing and defense. While LG AI Research has accumulated production and process application experience centered on the group's manufacturing affiliates, SK Telecom has also joined the competition by applying its independently developed sovereign AI foundation model to manufacturing sites.
SK Telecom said Wednesday it signed strategic memorandums of understanding (MOUs) with KG Steel, a steel manufacturer, and Konec, an auto parts manufacturer, to pursue on-site validation of AI agents based on its sovereign AI foundation model. It marks SK Telecom's first case of applying its independently developed sovereign AI foundation model to manufacturing.
Since April, SK Telecom has secured manufacturing site data held by KG Steel and Konec, including past process error and accident analysis reports, equipment manuals, and facility logs. Based on this, it developed a demo version of a "manufacturing-specialized AI agent" built on its sovereign AI foundation model "A.X K1." The developed demo version will be applied to actual processes in the second half of this year. KG Steel will deploy the AI agent on the cold rolling line at its Dangjin plant, which produces coated steel sheets, while Konec will deploy it in casting and machining processes, each providing related data to SK Telecom. Based on this data and on-site feedback, SK Telecom will advance the performance and inference speed of the manufacturing-specialized AI agent.
Not only SK Telecom but other companies participating in the sovereign AI model project are focusing on securing real-world industrial application cases ahead of the second evaluation in August. EXAONE, developed by LG AI Research, has already been deployed for production and process improvements centered on the group's manufacturing affiliates. LG Display trained domain knowledge of the OLED manufacturing process into an EXAONE-based AI production system. Through this, it detects anomalies occurring during operations in real time. According to LG, LG Display has generated cost savings of more than 200 billion won annually through this AI production system.
EXAONE has also been deployed in the finance and public sectors. With the London Stock Exchange Group (LSEG), LG developed "EXAONE Business Intelligence," a financial AI agent that predicts the returns of investment assets and provides commentary. LG CNS is pursuing EXAONE-based AX (AI transformation) projects in public-sector areas including the Ministry of Foreign Affairs, the National Police Agency, the Gyeonggi Provincial Office of Education, the Korean Intellectual Property Office, and the Ministry of the Interior and Safety.
Upstage, another participating company, is pursuing a strategy to expand its own AI foundation model "Solar WBL" into industrial fields such as finance, healthcare, manufacturing, and public services. In the manufacturing field, MakinaRocks, an industrial AI company participating in the consortium, is expected to play an important role. MakinaRocks holds validation experience in equipment anomaly detection accumulated on production lines such as those of Hyundai Motor.
The reason the sovereign AI model participants are pouring efforts into industrial site validation is that the axis of generative AI competition is shifting from general-purpose model performance to industry-specific specialized models. Whether problems can be solved using data at actual sites is a key variable in subsequent evaluations and commercialization competition. In particular, the manufacturing, defense, finance, and public sectors have high security requirements, and it is difficult to take site data outside. This is because process data, equipment manuals, accident histories, financial information, and administrative data are core assets of companies and institutions. In the case of manufacturing, the "knowledge isolation" problem, in which core know-how remains with specific skilled workers or departments, must also be solved. For this reason, analysts say a structure that closely connects with actual industrial sites to collect data and feeds it back into model advancement and on-site service improvement is essential.
Meanwhile, the four elite teams participating in the sovereign AI model project are conducting model development and advancement work ahead of the second-stage evaluation in August. The existing elite teams—LG AI Research, SK Telecom, and Upstage—must complete model development by the end of this month, while Motif Technologies, which joined in February, must complete development by the end of July. The four teams will undergo the stage evaluation in early August, after which three of them will advance to the next stage.
