The three major telecommunications companies in South Korea—SK Telecom, KT, and LG Uplus—are focusing on tokens, a key component of artificial intelligence (AI) operations. They are leveraging their connectivity and management capabilities to secure new revenue streams in the AI era.
On November 12, LG Uplus announced its plans to enter the AI token market next year, following a study on 'AI Token Optimization Recipes.' The company aims to target the market through partnerships with collaborators and by building its own infrastructure.
Tokens are the basic units of data consumed during the process of querying AI models and finding answers. As the use of AI expands and the number of inference models and AI agents increases, the demand for tokens has surged, prompting companies to seek efficiency in utilizing AI for their business purposes.
Telecom companies emphasize that their accumulated infrastructure capabilities from optimizing business processes can be applied to the AI token market. Lee Sang-yeop, Chief Technology Officer (CTO) of LG Uplus, stated, "Telecom companies have been optimizing in areas such as frequency and traffic, and this experience will contribute to the token business."
The CTO added, "We have validated token optimization in numerous practical services beyond theoretical efficiency, and we believe we can reduce the 'total cost' of tokens."
In the model routing phase, which connects to suitable AI models, LG Uplus has reduced the cost of processing general inquiries by 25% and coding agent costs by over 30%. In the coding area, where token consumption is highest, costs were cut by 90%. The company has also succeeded in lightweighting its AI models, reducing the size of its on-device AI model from 2.4 to 3 gigabytes (GB) to about 800 megabytes (MB).
LG Uplus has also introduced a Chip-Mix strategy, collaborating with NVIDIA and AMD for graphics processing units (GPUs) and validating neural processing units (NPUs) with domestic AI semiconductor company FuriosaAI. The CTO noted, "We plan to enhance efficiency by splitting and distributing tasks that require significant computation and bandwidth."
SK Telecom and KT are also launching token platforms to penetrate the market. SK Telecom is presenting a token gateway that aims to organically connect its core businesses in telecommunications, AI services, and AI infrastructure to dominate the token economy.
The company is particularly focused on managing customer AI usage patterns and connecting them to optimal AI models and infrastructure. SK Telecom aims to create a growth engine by consolidating its capabilities, including 23 million customers who can be linked to AI demand and a secure payment system.
KT recently unveiled its 'Token Factory,' which integrates and provides domestic and international AI models. This service allows customers to handle everything from subscription to cost management and payment in one place, offering models tailored to their needs.
KT has optimized its models using its self-developed AI model routing technology, 'Auto Model Router,' and has reduced cost burdens and response times through prompt compression technology. The company reported a 25% reduction in costs and has provided over 70 types of AI models since its launch last month, with plans to expand to more than 500 models in the future.
On November 12, LG Uplus announced its plans to enter the AI token market next year, following a study on 'AI Token Optimization Recipes.' The company aims to target the market through partnerships with collaborators and by building its own infrastructure.
Tokens are the basic units of data consumed during the process of querying AI models and finding answers. As the use of AI expands and the number of inference models and AI agents increases, the demand for tokens has surged, prompting companies to seek efficiency in utilizing AI for their business purposes.
Telecom companies emphasize that their accumulated infrastructure capabilities from optimizing business processes can be applied to the AI token market. Lee Sang-yeop, Chief Technology Officer (CTO) of LG Uplus, stated, "Telecom companies have been optimizing in areas such as frequency and traffic, and this experience will contribute to the token business."
The CTO added, "We have validated token optimization in numerous practical services beyond theoretical efficiency, and we believe we can reduce the 'total cost' of tokens."
In the model routing phase, which connects to suitable AI models, LG Uplus has reduced the cost of processing general inquiries by 25% and coding agent costs by over 30%. In the coding area, where token consumption is highest, costs were cut by 90%. The company has also succeeded in lightweighting its AI models, reducing the size of its on-device AI model from 2.4 to 3 gigabytes (GB) to about 800 megabytes (MB).
LG Uplus has also introduced a Chip-Mix strategy, collaborating with NVIDIA and AMD for graphics processing units (GPUs) and validating neural processing units (NPUs) with domestic AI semiconductor company FuriosaAI. The CTO noted, "We plan to enhance efficiency by splitting and distributing tasks that require significant computation and bandwidth."
SK Telecom and KT are also launching token platforms to penetrate the market. SK Telecom is presenting a token gateway that aims to organically connect its core businesses in telecommunications, AI services, and AI infrastructure to dominate the token economy.
The company is particularly focused on managing customer AI usage patterns and connecting them to optimal AI models and infrastructure. SK Telecom aims to create a growth engine by consolidating its capabilities, including 23 million customers who can be linked to AI demand and a secure payment system.
KT recently unveiled its 'Token Factory,' which integrates and provides domestic and international AI models. This service allows customers to handle everything from subscription to cost management and payment in one place, offering models tailored to their needs.
KT has optimized its models using its self-developed AI model routing technology, 'Auto Model Router,' and has reduced cost burdens and response times through prompt compression technology. The company reported a 25% reduction in costs and has provided over 70 types of AI models since its launch last month, with plans to expand to more than 500 models in the future.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.