Blab AI
SK Telecom to Pilot 519-Billion-Parameter AI Model in Steel and Auto-Parts Factories
SK Telecom (SKT) has inked memorandums of understanding with steelmaker KG Steel and automotive‑parts manufacturer KONEC to test its proprietary AI foundation model, A.X K1, in production facilities—marking the first time the model will run in the manufacturing sector.
A.X K1 is a 519‑billion‑parameter large‑language model that SKT unveiled earlier this year. Its design lets only about 33 billion parameters activate during each inference, enabling efficient operation on industrial hardware while preserving the reasoning power of a far larger system. Since April, SKT has been gathering process‑error reports, equipment manuals and operational log data from KG Steel and KONEC, using that information to build a demo version of a manufacturing‑specialized AI agent.
In the second half of the year, the partners plan to deploy the demo agent on KG Steel’s cold‑rolling line at its Dangjin plant, which produces coated steel sheets, and on KONEC’s die‑casting and machining lines. KG Steel’s output includes galvanized and cold‑rolled steel sheets, while KONEC supplies aluminum die‑cast components for automotive drivetrains to Tesla and major Korean automakers.
The pilot arrangement requires the manufacturers to share real‑time process data with SKT. In return, SKT will use that feedback to refine the agent’s performance and broaden its capabilities. Data collected during the pilot will also feed into training A.X K2, the next‑generation model currently under development.
SKT has noted that manufacturing has been a challenging domain for AI adoption because shop‑floor data is fragmented across departments and processes, and critical knowledge often remains with individual skilled workers rather than being systematically documented. The company also pointed out that many manufacturing sites are reluctant to adopt cloud‑based AI because it requires sending sensitive process data to external servers. A.X K1 supports on‑premise deployment, allowing the model to run inside a company’s own infrastructure without data leaving the facility.
The manufacturing push follows an MOU SKT signed last month with the Ministry of National Defense to apply the same model in defense applications. SKT said it plans to expand into finance, public services and health care.
The initiative is part of South Korea’s broader sovereign‑AI strategy, which seeks to reduce reliance on foreign technology and accelerate domestic AI development. By deploying A.X K1 in real‑world industrial settings, SKT aims to demonstrate the model’s ability to handle complex, high‑volume production data while addressing data‑security concerns that have slowed AI adoption in manufacturing.
At present, the pilot has not yet produced measurable performance metrics, but SKT expects the on‑premise agent to provide real‑time diagnostics, anomaly detection and predictive maintenance support. The company also plans to use the pilot data to fine‑tune the model for industry‑specific terminology and workflows.
In the coming months, SKT will monitor the agent’s impact on production efficiency, error rates and maintenance schedules. The company will also evaluate the feasibility of scaling the solution to other manufacturing sites and to additional sectors such as finance and healthcare.
The pilot represents a significant step toward integrating large‑language models into industrial operations, but challenges remain. The success of the deployment will depend on the quality of the data shared, the ability of the model to adapt to highly specialized processes, and the willingness of manufacturers to invest in on‑premise AI infrastructure.
SKT’s next‑generation model, A.X K2, will incorporate lessons learned from the pilot and is expected to feature further optimizations for industrial workloads. The company has not yet announced a release date for the new model.
Overall, the partnership between SKT, KG Steel and KONEC illustrates a growing trend of telecom operators leveraging their data‑processing capabilities to enter the industrial AI market, while addressing the unique data‑security and integration challenges that have historically limited AI adoption in manufacturing.
A.X K1 is a 519‑billion‑parameter large‑language model that SKT unveiled earlier this year. Its design lets only about 33 billion parameters activate during each inference, enabling efficient operation on industrial hardware while preserving the reasoning power of a far larger system. Since April, SKT has been gathering process‑error reports, equipment manuals and operational log data from KG Steel and KONEC, using that information to build a demo version of a manufacturing‑specialized AI agent.
In the second half of the year, the partners plan to deploy the demo agent on KG Steel’s cold‑rolling line at its Dangjin plant, which produces coated steel sheets, and on KONEC’s die‑casting and machining lines. KG Steel’s output includes galvanized and cold‑rolled steel sheets, while KONEC supplies aluminum die‑cast components for automotive drivetrains to Tesla and major Korean automakers.
The pilot arrangement requires the manufacturers to share real‑time process data with SKT. In return, SKT will use that feedback to refine the agent’s performance and broaden its capabilities. Data collected during the pilot will also feed into training A.X K2, the next‑generation model currently under development.
SKT has noted that manufacturing has been a challenging domain for AI adoption because shop‑floor data is fragmented across departments and processes, and critical knowledge often remains with individual skilled workers rather than being systematically documented. The company also pointed out that many manufacturing sites are reluctant to adopt cloud‑based AI because it requires sending sensitive process data to external servers. A.X K1 supports on‑premise deployment, allowing the model to run inside a company’s own infrastructure without data leaving the facility.
The manufacturing push follows an MOU SKT signed last month with the Ministry of National Defense to apply the same model in defense applications. SKT said it plans to expand into finance, public services and health care.
The initiative is part of South Korea’s broader sovereign‑AI strategy, which seeks to reduce reliance on foreign technology and accelerate domestic AI development. By deploying A.X K1 in real‑world industrial settings, SKT aims to demonstrate the model’s ability to handle complex, high‑volume production data while addressing data‑security concerns that have slowed AI adoption in manufacturing.
At present, the pilot has not yet produced measurable performance metrics, but SKT expects the on‑premise agent to provide real‑time diagnostics, anomaly detection and predictive maintenance support. The company also plans to use the pilot data to fine‑tune the model for industry‑specific terminology and workflows.
In the coming months, SKT will monitor the agent’s impact on production efficiency, error rates and maintenance schedules. The company will also evaluate the feasibility of scaling the solution to other manufacturing sites and to additional sectors such as finance and healthcare.
The pilot represents a significant step toward integrating large‑language models into industrial operations, but challenges remain. The success of the deployment will depend on the quality of the data shared, the ability of the model to adapt to highly specialized processes, and the willingness of manufacturers to invest in on‑premise AI infrastructure.
SKT’s next‑generation model, A.X K2, will incorporate lessons learned from the pilot and is expected to feature further optimizations for industrial workloads. The company has not yet announced a release date for the new model.
Overall, the partnership between SKT, KG Steel and KONEC illustrates a growing trend of telecom operators leveraging their data‑processing capabilities to enter the industrial AI market, while addressing the unique data‑security and integration challenges that have historically limited AI adoption in manufacturing.
