LG Uplus to Build AI That Uses Less Power and Responds Faster Through Token Optimization

[Financial News] LG Uplus is teaming up with a company specializing in artificial intelligence (AI) model optimization to secure token optimization technology that will improve AI operating efficiency. The goal is to reduce the graphics processing units (GPUs) and power required for AI services while improving response times and service quality.
LG Uplus announced on the 2nd that it is working with OptAI to secure token optimization technology that will improve AI operating efficiency. A token is the basic unit of data that AI processes when understanding a user's question and generating a response. The two companies plan to improve processing efficiency by making AI models lighter and enhancing their computational processes.
Drawing on its experience operating real-world AI services, LG Uplus will be responsible for validating and applying the technology. OptAI will research and develop technologies that enable AI models to run more lightly and quickly. The goal is to reduce GPU and power consumption while improving response times and service quality.
The two companies previously collaborated in the field of on-device AI, which runs AI directly on devices such as smartphones. They applied lightweighting technology to an EXAONE-based small language model (sLM), enabling it to run on the neural processing unit (NPU), a dedicated AI semiconductor in smartphones. According to LG Uplus, power consumption was reduced by 78% and the model size by 82% while maintaining performance at the same level as the existing central processing unit (CPU)-based approach.
This time, the scope of their joint research is expanding to server GPU environments. In ongoing research, the companies have improved the computational process through which AI generates responses to better suit real-world service environments. As a result, the amount of tokens that can be processed on the same GPU has increased by up to four times compared with the previous level. LG Uplus plans to apply the technology it secures in phases to its own AI services and large-scale AI infrastructure operations.
The two companies also jointly hosted the Efficient AI Tech Seminar on the 1st. Experts from industry, academia and research organizations, including Furiosa AI, VESSL AI, Korea Electronics Technology Institute (KETI) and Hanyang University, participated in the event and shared insights into AI model optimization, semiconductors and real-world service applications.
Sang-yeop Lee, chief technology officer (CTO) of LG Uplus, said, "AI competitiveness depends not simply on improving performance, but on how efficiently we can process many tokens using the same resources. We will expand research into token optimization technology by collaborating with various partners, including OptAI, and work to produce results that can be applied to real-world services."
[email protected] Yoon Hong-jip Reporter