South Korea Launches Nationwide Free AI Program with No Usage Limits
The South Korean Ministry of Science and ICT has selected three consortia led by SK Telecom, KT, and Kakao to operate the "AI for All" service for all citizens; general chatbots and public service AI agents will be provided to the public for free, with no usage limits for normal use.
The three consortia were chosen from six applicant teams. The government plans to sign contracts with the operators in September, with Kakao's beta version set to launch in September and SK Telecom's beta version in October, aiming for all three services to officially go live within the year.
This year, the government is providing computational support of up to 512 NVIDIA B200 GPUs for the three operators to use for development and initial launch; these GPUs will be provided in an online, Kubernetes-managed cluster and will not be distributed as physical chips to each consortium, with specific allocation numbers yet to be disclosed.
Starting in 2027, the South Korean government plans to provide budget support for the operational costs of nationwide services, but public materials have not disclosed the total project budget, subsidy ratios, duration, or the capital expenditures each of the three operators will bear. The so-called "unlimited" usage applies to normal use by regular users, but operators may still implement technical controls on concurrency, rate, and abnormal automated requests.
The product path is not limited to general conversation. SK Telecom plans to develop an agent using the A.X K2 model that can receive commands like "help me book a restaurant" and complete tasks, with partners including T Map Mobility, Shinhan Card, and 19 other institutions; Kakao plans to integrate the AI chat entry into KakaoTalk and advance phone-based AI services with LG Uplus; KT plans to embed functions into the Daum portal and Genie TV, reducing the need for separate app installations. sedaily+1
The services will cover high-frequency scenarios such as housing, healthcare, education, employment, and government services, extending to actual processes like appointments, applications, and payments. Project rules require operators to provide general chat, public agents, and specialized agents using both their own and other companies' models; reports indicate that local models must constitute the main part of the service, while foreign models can only be used when necessary and will not receive related government subsidies.
In terms of market mechanisms, the South Korean government will bear the initial customer acquisition and part of the inference costs through public budgets and GPU clusters, directing national users to local models, communication networks, payment, maps, portals, and government service entrances. Beneficiaries include SK Telecom, KT, Kakao, local foundational model developers, GPU cloud services, government software, and financial and life service platforms that can access agents; international model vendors will face pressure from policy biases affecting local traffic, default entrances, and subsidy eligibility. The real competition is not about "who provides free chat," but about who can establish long-term account relationships through user authorization, task execution, payments, and public service interfaces.
Source: Public Information
ABAB AI Insight
South Korea has previously established an online service foundation for residents through its digital government: tax, identity verification, public service applications, and payments have long been handled by various portals, operators, and financial institutions. "AI for All" is not about building chat products from scratch, but rather repackaging existing communication entrances, KakaoTalk social entrances, Daum content entrances, Genie TV home entrances, and public service processes into natural language task interfaces. SK Telecom's inclusion of T Map Mobility and Shinhan Card in its alliance, and Kakao's integration of the entrance into KakaoTalk, indicate that the three parties are competing for the user life service chain, not just for large model benchmark rankings.
The capital path shows a division of labor where the state provides the foundation and enterprises provide the entrance. The government reduces the threshold for model development and cold start with a centralized computational cluster of 512 B200s, and will subsidize nationwide service costs starting in 2027; operators will need to invest in their own models, application distribution, identity verification, risk control, customer service, and cooperative ecosystems. Since general chat and public agents must be free, commercialization is more likely to shift towards enterprise-specific agents, transaction conversion, payment services, cloud deployment, and high-value industry solutions, rather than directly selling tokens to the public.
Historically, South Korea's expansion of high-speed broadband, mobile communication, and e-government has relied on a combination of public infrastructure, consortium execution, and large-scale national access; China's WeChat embeds payment, mini-programs, and life services into social entrances, while Singapore connects government services with the unified identity system of Singpass. South Korea is currently in the "entrance integration" stage: its goal is not to surpass ChatGPT globally with a single model, but to allow local models to gain irreplaceable task data and high-frequency usage scenarios through local language, identity, payment, communication, and government service interfaces.
Essentially, this is about capital concentration: inference power, government budgets, user entrances, and public service interfaces are concentrating in a few selected consortia and local model ecosystems. Free services lower user migration costs, but also make operators more dependent on policy subsidies, scale effects, and task closures to cover inference costs; once AI agents can make appointments, applications, payments, and inquiries for users, the party controlling the default entrance will not only allocate information flow but also transaction flow, customer relationships, and behavioral data. Policy settings on local model ratios essentially use public demand to establish scale curves and continuous training data sources for domestic models.
ABAB News · Cognitive Laws
Free services lower usage thresholds, and entrances determine value attribution
Models can be purchased externally, but task data cannot be imported
Subsidies purchase computational power, and interfaces shape industry sovereignty