0
SK Telecom announced that it is deploying its proprietary AI model across manufacturing floors. / Courtesy of SKT

SK Telecom is applying its independently developed AI foundation model to manufacturing environments. This enables factories producing steel and automotive parts to actively utilize AI while maintaining high security by keeping their operational data strictly internal.

On the 25th, SKT announced that it has signed strategic memoranda of understanding (MOUs) with steel manufacturer KG Steel and automotive parts manufacturer Konek to push forward field trials of an AI agent powered by its proprietary AI foundation model.

This collaboration marks the very first instance of SKT applying its proprietary AI foundation model to the manufacturing sector. Since April, SKT has secured high-quality data from KG Steel and Konek—including past process error and accident analysis reports, equipment manuals, and logs—to develop a demo version of a manufacturing-specialized AI agent based on its proprietary AI foundation model, A.X K1.

A.X K1 is a large language model built with 519 billion parameters. While it excels at handling complex tasks, it activates only about 33 billion parameters during inference. This architecture, which keeps the overall model comprehensive but activates only the necessary segments, allows for highly efficient deployment on industrial floors.

In the second half of the year, SKT, KG Steel, and Konek will run field trials by deploying the demo version of the manufacturing-specialized AI agent across the cold rolling line of KG Steel's Dangjin plant, which produces galvanized steel sheets, as well as Konek's casting and processing lines.

KG Steel and Konek will share a larger volume of high-quality manufacturing process data with SKT in real time. Based on this data and field feedback generated during the trials, SKT will improve the performance and inference speed of the specialized AI agent while expanding its features.

Furthermore, the companies plan to utilize the manufacturing field data secured during the trials to train the upcoming A.X K2 model, which is currently under development. Following the completion of the trials, they intend to review the commercialization and formal adoption of the manufacturing-specialized AI agent, with options to upgrade to subsequent versions of SKT’s proprietary AI foundation model if necessary.

Manufacturing has traditionally been considered a challenging sector for AI adoption due to the slow digitization of field data and the fact that data is generated and managed separately across different processes and departments. Another limitation stems from variations in work execution based on an individual operator's skill level and experience. This creates a "knowledge silo" where core operational expertise remains confined to specific veteran workers, raising concerns that a factory's competitive edge could weaken when those veterans retire or change jobs.

Consequently, expectations are high that transforming scattered data and the empirical knowledge of skilled workers into digital assets—and introducing an AI agent trained on this data to the manufacturing floor—will boost efficiency by enabling swift responses to process errors. In addition, because security across individual processes is paramount in manufacturing, adopting cloud-based AI, which requires data to be transmitted externally, has been difficult. SKT's proprietary AI foundation model addresses this by supporting not only cloud delivery but also closed, on-premise environments, allowing companies to leverage AI within internal servers without leaking manufacturing process data outside.

Prior to this initiative, SKT signed an MOU with the Ministry of National Defense last month to utilize its proprietary AI foundation model in the defense sector, agreeing to deploy the model in military domains where security and data sovereignty are critical.

The advantage of a proprietary model that can operate without exporting data becomes even more pronounced in sectors that demand both strict security and specialized expertise. Starting with defense and manufacturing, SKT plans to expand the application of its proprietary AI foundation model to various domestic industries, including finance, public services, and healthcare. The company expects that specialized AI tailored to each sector—created by combining industry-specific data with its proprietary model—will strongly support the AI transformation across the entire domestic industrial landscape.

"For manufacturing floors where security is critical, a proprietary AI foundation model that can be utilized without sending data outside serves as an effective solution," said Jeong Seok-geun, Head of SKT’s AI CIC. "Beginning with our collaboration with KG Steel and Konek, we will accelerate the AI transformation of the manufacturing industry and continue to expand the use cases for our proprietary AI foundation model."

                                                                                                               Lee Ji-sun