[Edaily Reporter Shin Yeong-bin ] LG Uplus(032640)is set to enter the business-to-business (B2B) market next year, leveraging its “token optimization” technology designed to reduce the operating costs of artificial intelligence (AI) services.

LG Uplus held an “AI Token Optimization Study” for reporters on the 12th and unveiled its achievements in developing the related technology and its business plans. Starting next year, the company plans to pursue its enterprise AI token business by simultaneously forming partnerships with collaborators and building its own infrastructure.

A “token” refers to the basic unit used by AI to process text and other data. Costs are incurred based on usage. In particular, AI agents can consume more tokens than traditional chatbots because they repeatedly perform tasks such as searching, analyzing, and reasoning.

LG Uplus has focused not only on lowering the price per token but also on reducing the total cost required for AI to complete a single task.

The core technology is “model routing,” which automatically selects the most suitable AI model based on the task’s difficulty. This approach reduces unnecessary costs by using high-performance models for complex tasks and lightweight models for simple ones.

LG Uplus reported that internal testing showed a reduction of more than 25% in the cost of handling general inquiries and more than 30% in the cost of coding agents. The company also documented a case where a coding agent using its proprietary AI model reduced operating costs by more than 90% compared to the previously used model.

The company also achieved results in AI model lightweighting and improving inference efficiency. It reduced the size of its proprietary on-device AI model from the previous 2.4–3 GB to approximately 800 MB, while more than doubling the token generation speed. The company explained that by improving the computational processing method, it also increased processing efficiency by approximately fourfold on the same infrastructure. Two papers detailing these research achievements were accepted at the world-renowned AI academic conference “NeurIPS.”

The company is also pursuing a “chip mix” strategy to optimize semiconductor utilization. The plan is to reduce infrastructure costs by leveraging graphics processing units (GPUs) from NVIDIA and AMD, as well as neural processing units (NPUs) from FuriosaAI, according to specific task requirements. LG Uplus has been collaborating with FuriosaAI for over two years to validate the potential of NPUs.

LG Uplus plans to secure corporate customers by leveraging the operational experience it has accumulated through its AI call service “Exio,” its AI Customer Contact Center (AICC), and its internal AI transformation (AX) process. It is also taking the lead in operating AI models and advancing related technologies for “Everyone’s AI,” a collaborative service with Kakao set for commercial launch by the end of the year.

With KTCorporation launching its enterprise-grade “Token Factory” and SKTelecom also pushing forward with its “Token Gateway” business, competition in the AI token market among the three major telecom companies appears to be heating up in earnest.

Lee Sang-yeop, Chief Technology Officer (CTO) of LG Uplus, stated, “Dominance in the AI market has shifted beyond who can build the largest model to how precisely and cost-effectively one can deliver the business results customers desire.” He added, “The optimized technological capabilities we’ve honed through intense real-world challenges—such as AI for phone calls and in-house AX—will serve as the core moat of a sustainable AI business that no one can easily replicate.”