South Korea's government-led nationwide artificial intelligence (AI) service "AI for All" will make its first public appearance through beta testing this October. Three consortia—SK Telecom, KT, and Kakao—are preparing to unveil AI agents with distinct strategies, each leveraging their respective user touchpoints, with the goal of securing 15 million users alongside an official launch by year-end.
The Ministry of Science and ICT finalized the selection of these three consortia as project operators in August, following a public call for proposals in July. AI for All is a public initiative designed to give all South Korean citizens access to general-purpose AI chatbots and public AI agents at no cost. Core features will be offered free of charge, while companies may apply their own revenue models to advanced features they develop independently.
A key driver behind the project is the reality that approximately 21 million of South Korea's roughly 23 million generative AI users remain on free services. Foreign free AI services impose usage limits, and user conditions can shift unpredictably due to operator policy changes or subscription price hikes—concerns that shaped the government's rationale. The plan is to reduce dependence on foreign services by providing stable, free offerings built on South Korean AI models.
To support this, the government will initially provide 512 units of Nvidia's latest B200 graphics processing units (GPUs) to each consortium. Depending on performance, additional GPU infrastructure and operational cost support may follow. In return, consortia must use South Korean AI models for at least 50% of total usage, including 30% from other domestic companies' models.
Three Companies, Three Different Paths
The three consortia are preparing differentiated services based on their respective user touchpoints.
SK Telecom has adopted an "Appless" strategy that allows users to access AI without installing or launching a separate app. By using phone calls and text messaging—the telecom operator's core business—as the user interface, the company aims to improve accessibility for seniors and digitally underserved populations. At the heart of this approach is "action-oriented AI" that handles everything from planning to execution to result verification in a single flow upon user request.
A flagship feature is "Call on Your Behalf," where the AI directly places phone calls to stores or government offices that lack official APIs. The AI verifies information over the phone that is difficult to obtain through search alone, such as wait times at after-hours clinics or corkage fees at restaurants. A method allowing users to call a dedicated number and make requests directly to the AI is also in development, and the service is designed to be accessible even to non-SK Telecom subscribers.
Kim Ji-hoon, head of SK Telecom's AI Business Division, said, "We're not trying to compete head-to-head with global companies. We're focused on better addressing the pain points South Korean users have experienced with existing AI services."
Kakao is leveraging KakaoTalk as the primary entry point for its AI service. The goal is to make AI use as simple and habitual as chatting with a friend. The company plans to integrate a general-purpose chatbot with AI agents, enabling reservations, applications, and payments to be completed entirely within the messenger. LG Group is responsible for AI model development and deployment, with LG Uplus participating in model and infrastructure operations.
For public services, Kakao is developing a feature that identifies policies matching a user's eligibility criteria and then handles identity verification, public document issuance, and submission in a single flow. A representative scenario involves university students: the AI guides them through scholarship types and eligibility requirements, then handles everything from preparing supporting documents to submitting applications and receiving scholarship funds.
Kim Se-woong, vice president at Kakao, said, "What we're aiming for goes beyond just providing guidance. We want to connect the process all the way through to the point where public documents or scholarship funds actually reach the user."
KT is introducing "Ieum" (meaning "connection"), which links information search, comparison and judgment, and public service execution within a single conversation. The foundation is an "Ieum Inside" strategy that places AI entry points across partner channels such as portal Daum, Danawa, Musinsa, and Zigbang, as well as KT's telecom and IPTV services—allowing users to naturally invoke AI from services they already use. The structure's ability to onboard new brands at any time is cited as a key strength in scalability.
Behind the scenes, KT distributes workloads across multiple South Korean AI models—Upstage's Solar, Motif Technologies' Motif, Saltlux's Lucia, NC AI's Varco, and KT's own Midm—assigning tasks based on each model's strengths in reasoning, long-context processing, or multimodal capabilities.
Lee Jin-hyung, head of KT's AX Business Division, said, "We will create new AI services and connect them with existing services to broaden touchpoints. We'll focus on personalization that proactively suggests needed services based on the user's situation and life stage."
15 Million Users Targeted by Year-End
The three consortia plan to launch beta services in October, followed by official service launches in December. During the beta phase, SK Telecom and Kakao will begin with basic features such as chatbots, while KT will validate its service centered on Ieum Inside. AI agent features that perform actual tasks such as reservations and applications will be expanded in phases following stability checks.
Starting in 2027, a budget of 250 billion won (approximately $184 million) will be deployed. Of this, 170 billion won (approximately $125 million) will go toward securing 2,000 B200 GPUs, 30 billion won (approximately $22 million) will be directly allocated at 10 billion won (approximately $7.4 million) per consortium, and the remaining 50 billion won (approximately $37 million) will be used to foster a specialized agent ecosystem. The long-term goal is to evolve the service beyond simple information provision into a "one AI agent per citizen" model that actually executes various administrative and daily-life services.
The Shadow of Meta's Muse
However, some observers note that the domestic initiative is falling behind in the speed race. Meta's recently launched action-oriented AI agent "Muse" recorded 5.8 million mobile app downloads in the United States and Canada within just 20 days of release. Its standout feature is going beyond answering questions to checking users' emails and schedules and handling reservations and purchases on websites on their behalf. Meta provides 100 million tokens per user each month, making the service effectively free.
With Muse already securing actual users, concerns are being raised that the domestic service—which in its early stages is weighted toward search-style chatbots and safety verification—could see the gap widen with Big Tech players that have already gone all-in on action-oriented agents. While domestic platforms such as portals, messengers, and shopping services are cited as competitive advantages, the prevailing outlook is that as foreign AI services spread, collaboration between domestic platforms and overseas AI companies will actually expand.
Hwang Yong-seok, a professor at Konkuk University, noted, "The success of AI for All must be aligned with the industrial ecosystem. Right now, it appears to be a technology-driven, supply-centric project approached from a universal welfare perspective." He added, "We need to move in a direction that enhances platform competitiveness to benefit our industrial ecosystem and establishes defensive zones to avoid dependence on foreign services."
Deputy Prime Minister and Minister of Science and ICT Bae Kyung-hoon emphasized at a kickoff meeting on the 4th, "South Koreans are already using foreign AI services, so if we cannot create a clear differentiator compared to foreign offerings, AI for All will have failed. Let us all work together to deliver services that make citizens say 'wow.'"
Meanwhile, for AI to utilize citizens' administrative information—such as academic records and income data—to handle applications for various support programs on their behalf, institutional reforms regarding the scope of personal data linkage and private operators' information access rights remain prerequisite tasks. The extent to which administrative information can be utilized in the process of combining public services with private AI is expected to be a key point of discussion going forward.
Source: https://finance.biggo.com/news/5cf53815-3529-40a5-8fb5-6a198c2a9150
