LG’s Self-Evolving AI Strategy: K-EXAONE and the Rise of ‘Physical AI’

LG Group is shifting its artificial intelligence (AI) strategy towards “self-evolving, practical AI,” aiming to move beyond the competition for sheer AI model size and computational power. This new direction, unveiled at Mobile World Congress (MWC) 2026 in Barcelona, centers on delivering tangible value and solving real-world problems with its flagship AI model, K-EXAONE.

K-EXAONE: A Focus on Real-World Application

LG AI Research and LG Uplus jointly announced the roadmap for K-EXAONE, Korea’s leading AI model. Lim Woo-hyung, Co-Director of LG AI Research, emphasized that LG’s AI development isn’t about achieving the highest intelligence, but about creating a partner that assists human life in practical ways. The company aims to complete the highest-performing language model among existing global open-weight models within the first half of 2026. Korea Times

Building the Infrastructure: The Paju AI Data Center

To support K-EXAONE and its broader AI initiatives, LG Uplus is preparing the Paju AI Data Center (AIDC), slated for completion in 2027. This data center, located in the Seoul metropolitan area, will be the largest of its kind in the region, boasting a 200-megawatt scale and the capacity to house up to 120,000 graphics processing units (GPUs). Sedaily The AIDC is expected to consolidate capabilities across LG affiliates, including LG Electronics, LG Energy Solution, and LG CNS.

Ixio: LG Uplus’s Agentic AI Strategy

LG Uplus is focusing on its agentic AI service, ‘Ixio,’ aiming to become an “Ambient AI agent” connecting all devices. The company intends to evolve beyond being a network provider and become a software provider, integrating Ixio’s voice AI into various physical devices like cars, glasses, and humanoid robots. Biggo

Addressing Key Challenges in Agentic AI

LG Uplus acknowledges challenges in achieving successful agentic AI implementation, including maintaining continuous performance, domain optimization and cost, accountability for AI actions, and a lack of in-house infrastructure. Currently, less than 5% of companies adopting AI have achieved a return on investment (ROI). Biggo

Self-Evolving System and Trusted Orchestration

LG Uplus has developed a “Self-Evolving” system that can reflect frequently changing corporate policies and work standards into its AI model within a week – a significant improvement over the previous nine-month process. They’ve also introduced a “Trusted Orchestration” system to track and manage AI’s decision-making process, minimizing liability risks by managing the agent’s authority and operational history. Biggo

Hybrid AI Infrastructure and Cost Optimization

LG Uplus is pursuing a “hybrid AI infrastructure” to support high-performance physical AI, combining network and edge servers to overcome the limitations of on-device computing. The company is testing neural processing units (NPUs), including those from Furiosa AI, and conducting research and development on hybrid architectures. Addressing cost concerns, particularly the high token costs associated with voice-to-voice (V2V) models (more than 30 times higher than text-based models), is a priority for commercialization. Biggo

AI Native Network and 5G Standalone

To further support its AI initiatives, LG Uplus is establishing an “AI native network” based on the commercialization of 5G standalone mode (SA). This aims to provide a significantly lower latency environment for real-time interactions between voice AI and robots. Biggo

LG Uplus envisions a future where Ixio is integrated into various agents, transforming the company from a network-centered business to an AI software company focused on solving customer problems.