KT 拟将“AI for All”联盟从16家扩至20余家并自购GPU，属于无金额的业务规划性表态

KT is expanding the "AI for All" consortium it formed with 16 companies including Zigbang, Daum and Musinsa. The carrier said it has secured its own graphics processing units to keep the service running reliably, and that it will consider charging for business-use or premium public services once they launch.
KT said on the 12th that it is reviewing plans to expand the consortium from 16 companies to more than 20. Lee Jin-hyung, head of KT's AI transformation business division, said seven additional companies have asked to join, including major domestic financial groups and hotel and restaurant reservation service providers. "We plan to review additional partnerships with these companies and will continue to accept applications from others," Lee said.
Current participants in KT's consortium include Upstage, Saltlux, Daum, Musinsa, EBS and BC Card. The company appears to have concluded that more partners are needed to offer a wider range of services. KT has previously outlined a strategy of spreading AI quickly across services and industries through "Ieum Inside," which links a general-purpose AI agent based on real-time search with AI services specialized for more than 10 fields including public administration, education, finance and shopping.
"We will build the AI for All application into a core platform for public use of AI, and use the Ieum Inside strategy to apply AI for All features to the Daum search portal, to the various service touchpoints of consortium member companies and to KT's Genie TV," a KT official said. "Users will be able to enjoy the same AI experience naturally within the services they already use, without having to install a separate app."
For example, if Ieum Inside is applied to Zigbang, a consortium member, the AI would deliver messages such as: "A new recorded sale price has come in for the apartment you are viewing," "Of the 11 listings in this complex, two are below market price," and "It is 50 million won below yesterday's recorded sale price and was listed only two days ago."
In the public sector, KT plans to build AI agent services tailored to each user's situation, going beyond simple question-and-answer to offer proactive alerts and personalized recommendations of public benefits. Public alerts, public benefits matched to a user's age and place of residence, conversational guidance through procedures, and support for issuing and applying for documents are all expected to run within the Ieum app and website. "Going forward, we will expand the service into more areas such as farming and startups, and draw new companies into the business to upgrade it into an AI agent service that directly supports people's daily lives," a KT official said.
These services will be built on Token Factory, KT's AI operations platform. Various domestically developed AI models including Solar, Motif and Mi:dm, along with Rebellions' neural processing units and infrastructure optimization technology from Lablup and Vessl AI, will be operated in an integrated manner on Token Factory, automatically connecting the best-suited AI model for each service and delivering stable AI services.
Token Factory separates the processing of user prompts from response generation and allocates resources optimized for each stage, improving tokens per minute by about 31%. It can also cut input tokens by up to 80% through various prompt compression technologies, raising GPU utilization.
"To be genuinely committed to AI for All, KT has secured additional GPUs beyond those provided by the government," Lee said. "We technically hold a buffer of about 20% in case traffic surges, and we also have GPUs on hand for when demand exceeds that."
KT also left open the possibility of charging for some AI for All services. "If at some point the government opens a monetization model to the general public as well, we plan to separate out premium public services and provide them separately," Lee said. "We will also launch a PC version of an AI agent for professionals, especially for work use, in January as part of AI for All, and we have plans to apply a paid model to those services if needed."