South Korea’s Ministry of Science and ICT opened bidding on Monday for what would be the most ambitious deployment of government-funded artificial intelligence in history: a free, unlimited AI chatbot and public-service assistant available to all 52 million of the country’s residents, backed by domestic models and a statutory mandate to close a widening digital equity gap.

The project’s architecture raises a question that no other government has had to answer before: when a democracy commits to treating AI access as a basic public right, and simultaneously requires that at least half the AI system be built on homegrown models targeting 95% of frontier performance, does it risk creating a two-tier outcome — universal access on paper, frontier capability only for those who can afford to pay for it separately?

The Ministry of Science and ICT said applications for the “AI for Everyone” project will remain open through Aug. 11. The ministry plans to select two or three private-sector firms with proven consumer-service experience to build and operate the system. Following a beta launch in late September, the full service — a general-purpose AI chatbot paired with a more capable public-service agent — is expected to be live for all citizens before the end of the year.

South Korea’s AI Equity Gap, Quantified

The program is rooted in a stark government survey finding: just 31.9% of people classified as digitally vulnerable — including older adults and low-income residents — had experience using AI services, compared to 59.4% of the general public. That 27.5-percentage-point gap is not merely a convenience gap. As AI becomes embedded in employment applications, educational tools, government benefit navigation, and healthcare access, the divide compounds with each passing month into a structural disadvantage.

Deputy Prime Minister and Science and ICT Minister Bae Kyung-hoon has been direct about the stakes. “Problems such as job losses and wealth concentration due to AI are inevitable,” he said at a press conference marking the government’s first year in office in May. “AI for Everyone will create equal opportunities for everyone.” Bae has invoked the historical analogy of calculators and computers — technologies that became basic tools of economic participation in earlier eras — describing the proposed service as “more than a single service” that “will help people work, learn and live their daily lives together with AI.”

Research on digital equity broadly supports this concern. Studies of prior digital transitions consistently show that age, income, and geographic location predict technology adoption, and that the compounding effects of being left behind each successive wave — internet, smartphone, broadband — are cumulative. A Korea Research Institute for Human Settlements analysis published in May 2026 found that the regional gap in AI adoption capacity is larger than in any other technology domain, with the Seoul metropolitan area capturing AI development at a disproportionate rate.

How the System Is Built — and What It Will Actually Run On

The platform has two tiers. The first — a general-purpose AI chatbot — will launch in beta for the public in late September and remain free with no usage limits. The second, a more sophisticated public-service agent, is planned for later in 2026 and into 2027: it will proactively identify government benefits and administrative programs relevant to individual users, notify them in advance, and assist them with applications directly through the interface — replacing the need to navigate multiple government portals or call service centers.

The engineering constraint at launch is significant and worth understanding. The Ministry of Science and ICT will provide the selected companies with a combined 512 Nvidia B200 GPUs to power the initial service. A single B200 delivers approximately 9 petaFLOPS of FP8 compute. Running a 7-billion-parameter model — roughly the scale of South Korea’s smallest sovereign language models — a 512-GPU cluster can serve on the order of 2,500 to 7,500 simultaneous chatbot responses. For a country of 52 million people, that figure signals a beta and pilot phase, not national-scale deployment from day one. The full national-scale infrastructure will draw on South Korea’s broader GPU ecosystem: the Ministry’s directly purchased fleet of 13,000 GPUs (targeting 50,000 by 2030), plus the combined resources of Naver Cloud, NHN Cloud, and Kakao — the three operators running the country’s National AI Computing Center — whose total allocation exceeds 7,000 B200 units from a prior government contract.

Beginning in 2027, the government plans to provide direct budget support to cover the cost of running the service nationwide. The service is committed free through 2028, with private-sector co-investment discussed for subsequent years.

The Domestic Model Mandate — and What It Means for Performance

The most consequential technical requirement in the procurement is a mandate that developers use South Korean AI foundation models, meeting the Ministry’s own independent standards, for at least 50% of the system. This is not a soft preference — it is a structural procurement condition. In practice, it means the selected platform operators must route at least half of all AI inference through models from the five Korean AI consortia that the Ministry selected for its sovereign foundation model program: Naver Cloud (HyperCLOVA X), LG AI Research (Exaone), SK Telecom (A.X series), Upstage (Solar Pro), and NC AI (Varco).

These models are capable and specifically optimized for Korean language tasks. Naver’s HyperCLOVA X was trained on 6,500 times more Korean-language data than OpenAI’s GPT-4, according to the company, and outperforms GPT-4 on Korean-specific benchmarks. SK Telecom’s AX 3.1 Lite achieves approximately 96% of the top score on the KMMLU2 Korean language reasoning benchmark. The government’s stated performance target for the sovereign models as a class is to reach at least 95% of frontier model performance.

The honest caveat is parameter scale. Korea’s sovereign models currently operate in the 7-to-32-billion-parameter range — substantially smaller than the largest frontier commercial models, which operate at or beyond the 100-billion-parameter scale in terms of effective capability. Within Korean-language and Korean-cultural tasks, the performance gap may be modest. For tasks requiring broad world knowledge, complex multi-step reasoning, or multilingual queries, the gap may be more visible. Whether that gap matters for the specific use cases of a government public-service platform — answering benefit eligibility questions, navigating administrative procedures, providing civic information — is an open question that the beta launch will begin to answer.

This is the core tension the program must navigate: universal access paired with a mandated architecture that may not deliver the same quality as the frontier commercial AI that wealthier, more digitally fluent citizens will continue to use on top of it.

Built on Asia-Pacific’s First Comprehensive AI Law

South Korea’s “AI for Everyone” project does not exist in a policy vacuum. It sits within a nationally mandated legal and strategic framework that makes it the most structurally grounded citizen AI initiative any G20 government has attempted.

The Framework Act on the Development of Artificial Intelligence and Establishment of Trust — known as the AI Basic Act — took effect on January 22, 2026, making South Korea the second country globally to adopt comprehensive AI legislation, after the European Union. The law consolidates the work of 19 separate prior AI bills into a unified framework. It establishes a National AI Committee chaired by the president, an AI Safety Research Institute, risk-based obligations for high-impact AI systems in healthcare, employment, education, finance, and public safety, and mandatory disclosure requirements for generative AI.

The country’s AI Action Plan 2026–2028, finalized by the National AI Strategy Committee in February 2026 and spanning 99 execution tasks and more than 300 policy recommendations, includes “AI for Everyone” as a named instrument for what the government calls an “AI basic society” — one in which access to AI is treated as an essential right, on par with access to education or public health services. The service is to remain free under government funding through 2028.

Sovereign AI and the Nvidia Dependency Question

South Korea’s capacity to build this service rests on a substantial national AI infrastructure investment — but one that independent analysts note has a structural dependency at its core. The country’s 260,000-GPU national AI ecosystem announced at the APEC Summit in October 2025 combines government-procured hardware with private-sector investments from Samsung, SK Group, Hyundai Motor, and Naver — each deploying up to 50,000 to 60,000 Nvidia Blackwell GPUs. South Korea also ranked first globally in AI patents per capita and third globally in notable AI models released last year, with five models behind the United States (50) and China (30), according to the Stanford AI Index 2026 (14.31 patents per 100,000 people).

Critics have noted that placing Nvidia chips on Korean soil does not, by itself, constitute meaningful data sovereignty. Digital Today’s analysis of the national AI infrastructure concluded that “sovereign AI may not go beyond placing Nvidia chips on South Korean territory,” given that GPU manufacturing, CUDA software frameworks, and foundational model training infrastructure remain concentrated in US hands. South Korea’s export-control exposure to US chip policy is an additional layer of risk: the country experienced the practical consequences of supply-chain concentration when US export restrictions affected a portion of advanced chip imports in 2023.

The “AI for Everyone” project sidesteps part of this critique by defining sovereignty at the model layer — requiring domestic Korean foundation models to power at least half the system — rather than at the hardware layer. But the hardware dependency remains a real long-term variable for any sovereign AI policy built on Nvidia’s platform.

What 52 Million People Would Gain — and What Comes Next

For the 31.9% of digitally vulnerable South Korean residents who have not yet used AI in any form, the practical value of an unlimited, free, Korean-language AI agent is not abstract. Navigating government benefit eligibility — a process that currently requires identifying the relevant ministry, locating the correct portal, decoding bureaucratic language, and completing forms that often presuppose digital fluency — is precisely the class of task where a well-designed conversational AI agent can dramatically reduce friction. The public-service tier of the platform, targeted for 2027, is specifically designed to proactively surface benefit eligibility rather than waiting for citizens to ask.

The number of AI Digital Learning Centers in South Korea has nearly doubled from 37 to 69 since the current government took office, and the number of citizens participating in AI education programs has grown from 910,000 to 1.3 million. These programs represent the demand-side complement to the supply-side infrastructure of “AI for Everyone.”

The immediate milestone is the August 11 vendor application deadline, after which the Ministry will select its implementation partners and push toward the late-September beta launch. Whether South Korea’s model becomes a template for other governments — and whether the domestic model requirement proves to be a principled architectural constraint or a performance ceiling — will depend on execution. Several dozen countries are watching what South Korea does with the world’s most concrete commitment to AI as a public good.

Frequently Asked QuestionsWhat is South Korea’s “AI for Everyone” project, and who can use it?

“AI for Everyone” is a South Korean government initiative to provide every resident of the country with free, unlimited access to an AI chatbot and public-service assistant, built primarily on domestically developed Korean AI models. The Ministry of Science and ICT opened a competitive bidding process on July 14, 2026, to select two or three private companies to build and operate the service. A beta chatbot is planned for late September 2026, with a full service — including an AI agent that proactively identifies government benefits for individual users — targeted by year-end. The service will remain free through 2028 under government funding.

Why is South Korea the first G20 nation to do this?

South Korea has the regulatory infrastructure, national AI strategy, and domestic AI industry ecosystem that no other G20 government has yet assembled simultaneously. The AI Basic Act, which took effect in January 2026 as Asia-Pacific’s first comprehensive AI law, created a legal mandate for the government to actively promote AI access as a public benefit. The government has also invested in five competing domestic AI foundation model consortia (Naver, LG, SK Telecom, Upstage, NC AI), which created the model supply chain that makes the 50% domestic model requirement technically feasible. Other G20 governments are still at the policy-discussion or limited-pilot stage for population-scale free AI.

What does the requirement to use domestic Korean AI models actually mean for the service’s quality?

At least 50% of the “AI for Everyone” system must be built on South Korean AI foundation models meeting the Ministry’s independent standards — specifically the models developed by the five sovereign AI consortia. These models operate primarily in the 7-to-32-billion-parameter range and are specifically optimized for Korean language and cultural tasks, achieving approximately 95-96% of top benchmark scores in Korean-language reasoning. The government’s stated performance target is to reach 95% of frontier model performance overall. The honest assessment is that within Korean-language public-service tasks — the core use case — the gap may be manageable. For complex reasoning tasks or multilingual queries, some performance difference from frontier commercial models is likely. The beta launch in late September will provide the first real evidence of how wide that gap is in practice.

Could a free government AI create a new kind of inequality — between citizens who use it and those who pay for frontier commercial AI?

This is the central unanswered policy question the project raises. Universal free access addresses one dimension of the AI divide: the cost and availability barrier that currently locks out the 31.9% of digitally vulnerable South Koreans who have never used AI. But it leaves open a second dimension: quality. If the government-provided service delivers materially lower performance than frontier commercial AI on the kinds of tasks that matter most for employment, education, and financial decisions, then wealthy citizens will use both — the free public service for basic tasks and premium commercial services for high-stakes ones — while citizens who rely exclusively on the public service may receive systematically less capable AI assistance. Whether South Korea’s domestic models close enough of that gap to make the distinction negligible is the defining question of the program’s first year.