South Korea's Ministry of Science and ICT confirmed on August 28 which companies will build the government's "AI for All" chatbot, selecting three consortia led by SK Telecom, KT and Kakao to give all 52 million residents free, unlimited access to a domestically developed AI assistant. The selection, drawn from six bidders, turns an idea first floated in July into a concrete build-out: agreements are due to be signed in September, a public beta follows the same month, and a full national rollout is targeted before the end of the year. It is a scale of commitment no other government has yet matched — treating a general-purpose AI chatbot less like a tech product and more like a utility, alongside broadband and public schooling.
Each consortium brings a different distribution strategy and a different mix of domestic models. SK Telecom's group — which includes Shinhan Card, Hana Card, Tmap Mobility, SK Broadband and eleven other partners — is building an "action-oriented" agent that can complete tasks end to end, including placing phone calls to businesses or government offices on a user's behalf. It will run primarily on SK Telecom's own A.X K2 model, a 688-billion-parameter system developed under Korea's separate sovereign foundation model project, supplemented by Upstage's Solar Open 2, Motif Technologies' Motif 3 and LG AI Research's K-EXAONE 2.0. Kakao's consortium, joined by LG AI Research, Shinhan Bank, Seoul National University Hospital and others, will route users through KakaoTalk, the messaging app nearly every South Korean already uses, combining a general-purpose chatbot with agents that handle reservations, applications and payments using Kakao's Kanana and LG's EXAONE models. KT's group will connect its service to partner platforms including Daum, Musinsa and Zigbang, powered chiefly by its Mi:dm K Pro 2.5 model alongside Solar Pro 4 and Motif 3.
Notably absent is Naver, Korea's largest search and portal operator, which declined to bid. The company said it chose to focus on its existing AI business rather than commit to the project's tight timeline and its requirement to incorporate rival firms' models. Naver's HyperCLOVA X had already been eliminated from the government's national sovereign foundation model project in January over data-independence concerns, and executives were reportedly wary of a second setback. That earlier project — which in August 2025 selected Naver Cloud, Upstage, SK Telecom, NC AI and LG AI Research as Korea's "K-AI" champions with a combined 200 billion won in state support — effectively built the supply of homegrown models that AI for All now draws on, after Naver and NC AI were dropped in a later evaluation round.
The domestic-model mandate is the policy's structural core. Selected operators must route at least 50% of user queries through their own certified Korean foundation model and another 30% through models built by other domestic companies, pushing the floor for homegrown AI usage to 80% of total traffic. Foreign models such as ChatGPT, Gemini or Claude are permitted only in limited capacity, and any such usage falls outside the government's cost support — operators would have to pay for it themselves. The government is backing the effort with a pooled allocation of 512 Nvidia B200 GPUs across the three operators this year, and has pledged to begin subsidizing nationwide service costs from 2027, though the size of that funding has not been finalized.
Deputy Prime Minister and Science Minister Bae Kyung-hoon has framed the project as foundational infrastructure rather than a convenience app. "AI for All is more than just a service; it is the calculator and computer of the AI era," he said at the program's launch ceremony in Gwangju. Describing the roadmap beyond the initial chatbot, Bae said the ministry plans to build toward individualized service: "In addition to a general-purpose AI chatbot service, we will continue to expand public AI agent services," he told reporters during a July policy briefing with President Lee Jae-myung. "Starting in 2027, we will advance the service into a one-agent-for-every-citizen model."
Why It Matters
AI for All is the clearest test yet of a state treating a frontier AI assistant as public infrastructure rather than a commercial product. The sovereignty argument is straightforward: only about 1.8 million South Koreans currently pay for a commercial AI subscription even though roughly 23 million, or 44.5% of the population, already use generative AI — mostly through foreign services. Officials argue that leaving an entire population dependent on ChatGPT, Gemini or Claude cedes both data and strategic leverage to companies outside Korean jurisdiction, especially as the country's own AI Basic Act, in force since January, gives the government legal grounding to promote AI access as a public benefit.
Critics and analysts complicate that sovereignty framing on two fronts. First, the hardware underneath the "sovereign" models is still Nvidia silicon — Counterpoint Research has estimated Nvidia chips account for over 92% of global sovereign AI deployments — leading some observers to argue Korea's sovereignty claim really operates only at the model layer, not the infrastructure layer beneath it. Second, privacy advocates worry about what happens when a messaging platform used by nearly the entire country, and telecom and card companies with detailed behavioral data, become the default rails for a state-endorsed AI agent. Korea has direct precedent for the risk: the 2021 "Iruda" chatbot scandal, in which a startup trained a bot on roughly 9.4 billion unfiltered KakaoTalk messages without adequate user consent, still shapes debate over how AI systems handle personal data. Civic groups have separately pushed back this year against legislation that would ease restrictions on using identifiable personal data to train AI models, warning that "sandbox-style" flexibility could normalize opaque data practices. Supporters counter that a single, publicly accountable national service — built under an active regulatory framework — is more scrutinizable than the current default of citizens' data flowing to unregulated foreign platforms.
What to Watch
Watch for the September beta to reveal whether the 512-GPU allocation can actually support tens of millions of simultaneous users without degrading response quality, a concern industry analysts flagged even before consortia were chosen. Also worth tracking: how the government structures its 2027 subsidy program, whether advertising or another revenue model emerges to sustain "free and unlimited" access, and whether the privacy safeguards around KakaoTalk- and telecom-based distribution satisfy the civic groups already warning about data concentration at national scale.
“AI for All is more than just a service; it is the calculator and computer of the AI era.”— Bae Kyung-hoon, Minister of Science and ICT, South Korea