Key Takeaways

  • South Korea is providing free, uncapped generative AI access through public systems and widely used consumer applications.
  • Domestic models will handle at least 80% of subsidized queries, supporting Seoul’s push for technological sovereignty.
  • Up to 512 Nvidia B200 chips will support the service, but cost control, security and model accuracy remain significant operational tests.

South Korea is moving generative AI from a consumer subscription product toward something closer to national digital infrastructure. Its "AI for Everyone" initiative is designed to give roughly 52 million residents free access to AI chatbots and public-service assistants, with no token limits on usage.

The proposed services go beyond general-purpose question answering. Residents could use the systems to book medical appointments, seek help finding an apartment and obtain guidance about taxes. Small businesses could calculate tax obligations or check eligibility for government support, while parents could receive educational recommendations for their children.

Three technology consortia selected by Seoul will deliver the services. Each is expected to launch a dedicated application and integrate AI into products that South Koreans already use. Naver Cloud, NHN Cloud and Kakao are among the domestic companies underpinning the country’s national AI infrastructure, giving the program distribution channels as well as computing capacity.

That distribution piece matters. Kakao’s messaging platform, for example, already occupies a broad role in daily digital life. Putting a public AI assistant inside familiar applications could reduce the friction associated with asking residents to register for another government portal.

Beta testing starts in September, followed by a full rollout later in 2026. The government plans to allocate as many as 512 Nvidia B200 chips among the three groups and contribute toward operating expenses. The Ministry of Science and ICT has not disclosed the program’s total cost.

Providing "free" AI incurs a substantial marginal cost. Every query consumes computing resources, and uncapped access makes demand harder to predict. IDC projects global AI infrastructure spending at about $497 billion in 2026, with the market expected to surpass $1 trillion by 2029. South Korea’s service will therefore test whether a government can manage consumer-scale inference without introducing visible quotas or degraded performance.

The domestic model requirements are equally consequential. Operators are required to route at least half of queries through their own certified Korean foundation model and another 30% through models developed by other Korean companies. That creates an 80% floor for domestic processing. Foreign models can be used when the ministry considers them appropriate, but those queries will not receive subsidies.

It is a direct response to current market behavior. Roughly two-thirds of South Koreans have tried AI, and about 23 million regularly use generative tools. Ministry tender figures showed 23.45 million Korean ChatGPT users in April, compared with 8.45 million for Gemini and 2.41 million for Claude. Only around 1.8 million residents pay for AI services.

Can a state-funded alternative shift that entrenched usage? Price removes one barrier, but model quality, reliability and integration will probably determine whether people change habits.

President Lee Jae Myung’s administration has earmarked about $7.2 billion, or 10 trillion won, for AI spending in 2026, three times the previous year’s commitment. Separately, the Ministry of Science and ICT’s 2026 budget allocates about $3.5 billion to AI, a 30% year-on-year increase, according to the International Trade Administration. Prior supplementary funding included roughly $106 million for GPU resources and $34 million for data supporting domestic foundation models.

Security and accountability could become harder as the assistants connect to tax, housing and healthcare workflows. South Korea’s National Network Security Framework Guidelines v1.0, issued in 2025, provide layered requirements for cloud and AI use in public systems. Still, a chatbot offering tax guidance is not the same as one submitting a binding filing. Agencies will need clear boundaries around authentication, data retention, human review and responsibility for incorrect answers.

South Korea is part of a wider government shift. The OECD reports that 67% of governments across its membership use AI in public-service design and delivery. The European Commission’s InvestAI program, launched in February 2025, aims to mobilize $232 billion (€200 billion), while India’s $1.25 billion IndiaAI Mission has subsidized computing access and funded 20 domestic model projects.

Seoul’s distinction is its focus on the finished service rather than infrastructure alone. If adoption grows, the program could create a large domestic market for Korean-language models, cloud operations and specialized public-sector applications. It could also expose the practical limits of uncapped access. Either way, South Korea is turning sovereign AI from an industrial policy slogan into a service residents can actually use.