South Korea's telecom industry is paying attention to tokens, a new unit for charging in the AI era. It aims to expand billing and settlement capabilities built in telecoms into the AI market to secure new revenue sources.

As corporate adoption of AI expands, demand is growing to use multiple AI models efficiently and manage costs in an integrated way, the telecom industry said on Thursday. Telecom companies are therefore detailing AI business models that use existing billing systems and their assets, including corporate customer bases and data centres.

KT has introduced Token Factory, which integrates AI model use and cost management, and SK Telecom is also pushing a Token Gateway business. LG Uplus is also reviewing related businesses such as token pricing plans and has moved to secure technologies that improve AI processing efficiency.

The next unit after voice and data is tokens as telecom billing expands into AI

Tokens are the basic unit processed when AI understands questions, documents or code and generates answers. If existing telecom services charged based on call time and data usage, AI services use the number of tokens used for input and output as a key billing standard.

It is similar to telecoms in that fees are charged based on usage, but the billing structure is more complex. Token prices and processing performance differ by AI model, and even within the same model, different rates may apply to input and output. As companies use multiple AI models for work, tracking usage and fees by model and controlling budgets has emerged as a challenge.

Telecom companies are focusing on that demand. They aim to expand their existing business structure, which measured network usage and billed customers, into the area of AI cost management. Telecom companies already have systems to manage large volumes of usage and issue regular bills. The core plan is to combine authentication, payments and customer management to build a business model that covers AI service subscriptions, use and settlement.

Token Factory, which KT will launch on Oct. 30, is a service that integrates model selection, usage checks, cost management and payments while using domestic and overseas AI models on a single platform. KT plans to link more than 500 AI models through cooperation with global AI model distribution platform OpenRouter. Companies can choose and use multiple AI models on a single platform without subscribing or paying separately for each model.

SK Telecom presented a token gateway as a new business area in the AI era in its "SKT Growth Gear" strategy announced this month. "A token gateway is a business that supports customers to use AI at a more efficient cost by connecting various AI models and infrastructure to customer requests," an SK Telecom official said. "The goal is integrated management of which AI model to use, how much work to assign, how much was used and how much it cost."

Beyond linking models, a race to optimise costs linked to AI infrastructure

Competitiveness in the token business hinges on how much token usage and operating costs can be reduced while maintaining performance customers want.

KT points to its in-house "Auto Model Router" as a strength of Token Factory. Auto Model Router automatically selects a suitable AI model by analysing the type and difficulty of work, required quality and costs. It uses relatively inexpensive models for simple tasks and links high-performance models for complex analysis. It also combines prompt compression technology that reduces unnecessary expressions while maintaining the meaning of the question.

KT is also linking token distribution and infrastructure businesses, including providing AI inference services directly from domestic data centres. It plans to target corporate customers for whom data sovereignty and security are important by brokering overseas AI models while running AI models on its own infrastructure.

SK Telecom also cites model selection and cost optimisation as core functions of its token gateway. It aims to manage the entire process of AI performing work, beyond simple model linking or cost reduction. SK Telecom plans to link demand generated in AI services to its own models and data centres and combine that with telecom capabilities such as about 23 million customer touchpoints and identity verification, payments and billing.

LG Uplus is focusing on securing foundational technologies that increase token processing efficiency. Two AI token optimisation research papers involving company researchers were recently accepted at workshops at the AI academic conference NeurIPS 2026. LG Uplus applied model lightweighting and inference acceleration technologies to small language models to increase AI processing speed in a smartphone environment by about 4.5 times compared with the previous method and reduce battery consumption by about 77 percent.

The company is reviewing ways to apply the optimisation technologies secured through the research to its AI services and later expand them to server GPU environments. There is no detailed business blueprint yet, but there is analysis that technology that lowers AI processing costs could translate into competitiveness for the company in the future.

Even as token demand surges, profitability remains a challenge as competition heats up

Data disclosed by KT citing OpenRouter statistics showed AI token consumption on the platform rose about 35-fold over the past year. Token demand is expected to expand further if AI agents that automatically carry out complex tasks become widespread. Token usage can increase in the process of calling multiple AI models and verifying results.

Still, an increase in token consumption does not immediately translate into higher profits for telecom companies. "Brokerage models that link external AI models need to consider the costs paid to original model makers and margin settings," an AI industry official said. "Operating infrastructure will also cost a lot."

KT is also placing more weight on generating revenue using its own infrastructure than on simple token brokerage. Jin-hyung Lee (이진형), head of KT's AX Business Division, said at a press briefing on Oct. 27, "Simple brokerage fees are not the main business model." He said margins generated in the process of directly producing tokens using GPUs at KT data centres are the main revenue model.

Competition with global AI model suppliers and cloud operators is also a challenge to overcome. Amazon Web Services also integrates and provides various AI models through Amazon Bedrock. If telecom companies remain at the level of simply linking models and charging fees, it may be difficult to differentiate from existing players.

A telecom industry official said, "Telecom companies aiming to become AI companies have a strong character of first presenting new growth engines," adding, "They will need to prove specific cost reduction effects to secure customers."

Security and data management capabilities are also expected to become important. There is a risk of information leakage when a company's internal documents or customer information are transmitted to AI models. In particular, when overseas AI models and domestic data centres are used together, the industry diagnosis is that systems are needed to clearly manage data processing locations and access rights.