LG Uplus said on Thursday it revealed a platform-building case to stably run artificial intelligence models in real services at the '2026 Modern Agentic Application Day' event hosted by Amazon Web Services (AWS).

At the event, LG Uplus shared its experience of converting its existing on-premises-centred AI development environment into a flexible cloud-linked structure and of building a platform that efficiently connects the entire process from AI development to service operations.

The platform introduced this time focused on reducing the disconnect that had occurred between AI model development and service operations stages. It integrated into a single flow what had been separate processes for training, evaluation, deployment and operation, which had led to repetitive work during the transition to service.

LG Uplus designed the platform on the view that it is important to maintain 'Model Ready' status that can always be used immediately in services, rather than a structure that ends once AI is developed. It connected the full process from data collection to training, evaluation, deployment and operation in a single pipeline so developers and operators can handle AI under consistent standards and environments.

LG Uplus adopted a hybrid infrastructure architecture based on Amazon EKS, AWS' managed Kubernetes service. It integrated its own on-premises graphics processing unit infrastructure as hybrid nodes of an Amazon EKS cluster and ran the Kubernetes control plane, which centrally controls the entire cluster, as a fully managed AWS service to reduce the platform management burden.

The platform also features an improved method of using GPU resources. Instead of fixed allocation of GPUs by equipment unit as before, it applied a structure that flexibly distributes resources as needed. As a result, it became able to use GPU resources more efficiently when AI model training or service delivery is needed.

Kwon Ki-deok (권기덕), head of LG Uplus' AX Engineering Lab, said LG Uplus is strengthening engineering platform capabilities based on an AI-driven development lifecycle spanning data collection, model development, deployment and operation, and GPU operations. He said it will continue to raise AI service quality and operational stability through technical cooperation with AWS.