The new entry into the token market
LG Uplus has announced that starting next year it will enter the artificial intelligence token business, according to a report by ETNews. The company describes the initiative as a way to reduce the burden of AI transformation (AX) on enterprise customers by leveraging internally developed token optimization technologies. The news, reported on November 15, places the Korean group in a segment where other telecommunications operators are already active.
According to the material, the strategy focuses on the total cost of completing a task (TCO), not only the unit price of the token. CTO Lee Sang-yeop said that leadership in the AI market no longer depends on who builds larger models, but on how accurately and economically a provider can achieve the business outcomes the customer requires. He also indicated that the total cost needed to complete a task, more than the token price, will become the industry's technological moat.
The stated technical components
LG Uplus identifies three main elements: so-called harness engineering, which designs the workflows of AI agents; intelligent routing that assigns models based on the type of request; and a chip mix that combines different types of semiconductors to optimize inference. The idea is to prevent agents from repeating unnecessary searches or checks and to pair each task with the most suitable model, thereby reducing token consumption.
The material reports proof-of-concept cases in which the cost of processing generic queries was reduced by 25% and the cost for coding agents by more than 30%. The company also says it has verified a 90% cut in the operating costs of AI models in the coding area, where token consumption is high. It also cites an on-device model reduced to about one-third of the size with more than 95% of the performance of general cloud models, a token generation speed more than double thanks to a small model that generates and a large model that verifies, and processing efficiency about four times higher on the same infrastructure.
Infrastructure and competitive landscape
On the hardware side, the company announces a chip mix strategy that combines Nvidia and AMD GPUs with FuriosaAI's domestic NPU. The different units are used separately for the request analysis phase (prefill) and the response generation phase (decode), distributing compute-intensive workloads and memory-bandwidth-sensitive workloads to lower infrastructure costs. LG Uplus plans to combine partnerships with suppliers and direct buildout to target the B2B token market starting next year.
With this entry, all three major Korean telecommunications operators are now engaged in the token economy: KT had already launched the enterprise platform Token Factory for automatic model connection and billing management, while SK Telecom has listed Token Gateway among its new businesses. Still to be verified are the actual launch timelines, prices, commercial availability of the services and the actual replicability of the savings claimed in large-scale PoCs.
