[DigitalToday reporter Seulgi Son] The government will unveil 'AI for Everyone' later this year to provide all citizens with a free, unlimited AI service. The government says the service is needed to reduce reliance on foreign AI. Some say private-sector free AI is already widely used and that, with public GPUs being投入, AI for Everyone needs a differentiated role of its own.

Guidelines for the 'Support for universal use of AI services for all citizens (AI for Everyone)' project, posted on July 13 by the Ministry of Science and ICT and the National IT Industry Promotion Agency (NIPA), say the government will allocate up to 512 Nvidia B200 GPUs to 2 to 3 private companies. The selected companies must start a beta service by late September and, from December, provide a general-purpose AI chatbot with no limits on cost or usage.

It is not that the market lacks free AI. Monthly users in South Korea in April cited in the guidelines were 23.45 million for ChatGPT, 8.45 million for Google Gemini and 2.41 million for Anthropic Claude. These services offer free versions with certain limits on models and usage. In South Korea, services such as Wrtn also provide generative AI services for free.

The rate of AI use among the public is already substantial. In the National Information Society Agency (NIA) 2025 Survey on Internet Usage, the share of respondents with experience using AI services was 67 percent, and the share with experience using generative AI was 44.5 percent. The guidelines also assessed that most AI users in South Korea use free versions of foreign AI services.

The government says AI for Everyone is needed to reduce dependence on foreign AI. It says foreign free versions have usage limits and that prices or service terms could change in the future, making it difficult to control them domestically. The logic is that South Korea needs to secure stable homegrown services before AI becomes a tool needed for the public’s economic and social activities.

A Ministry of Science and ICT official said, "The first goal of the project is to provide an AI chatbot to all citizens in a stable manner." The official added, "If we provide performance comparable to foreign free versions without burdens of cost or usage limits, we believe we can draw users to domestic services."

Ultimately, the success of the project depends on whether domestic AI can be competitive enough to move users away from existing foreign free services.

Selected companies must keep the share of use of domestic models that meet the government’s own AI foundation model standards at 50 percent or more. Separately, they must also use domestic models developed by other companies for at least 30 percent. Foreign models can be used on a limited basis only for the minimum necessary functions, such as advanced reasoning and multimodal capabilities.

The government says domestic companies can deliver performance that can compete with foreign free versions. A Ministry of Science and ICT official said, "It is realistically difficult to catch up to the level of global paid services, but companies are showing confidence that they can deliver performance comparable to free versions."

But while the government has put forward a goal of matching foreign free AI, the guidelines did not present the foreign models that would serve as comparators or any common performance standards. The guidelines also have applicant companies propose ways to differentiate functions, technology and user experience compared with existing domestic and foreign services.

Some in the industry say the competitiveness of a nationwide AI service cannot be secured through GPU provision alone.

An AI industry official said, "A nationwide AI service is not built on model performance alone. Content, platforms and infrastructure to stably handle large-scale users must be supported together." The official added, "Users will not keep using an inconvenient service just because it is domestic or supported by the government. Investment is needed not only in GPUs but also in securing content and in service development and operations."

Another official said, "It is difficult to judge actual service competitiveness based only on benchmark scores for a model." The official added, "Only by comparing response accuracy, response speed, and error and outage rates under the same questions and usage environment can we confirm whether the quality is comparable to foreign free services."