The government has decided to select and provide focused support for an 'AI national representative' with the goal of developing world-class artificial intelligence (AI) models. To this end, it announced plans to nurture top-tier AI talent and expand infrastructure such as high-performance Graphics Processing Units (GPUs). However, the industry expresses concerns over the lack of specific implementation plans regarding budget and timing, fearing ineffectiveness.
◇ 'We will create a global LLM with the 'AI national representative''
According to the Ministry of Science and ICT, on the 26th, the National AI Committee announced the 'National AI Capability Enhancement Plan through the Expansion of AI Computing Infrastructure,' chaired by Acting President Choi Sang-mok, who is also the Minister of Strategy and Finance, on the 20th. The measures focus on rapidly expanding AI infrastructure, including cutting-edge GPUs, and increasing support for AI startups to incentivize talented individuals to engage in software development.
First, the government will promote the 'World Best LLM' project, a national representative initiative, within this year. It plans to select an 'elite team of AI national representatives' from corporations and to provide comprehensive support for research resources, such as data and GPUs, with the goal of developing a globally competitive large language model (LLM). The exact timing for the 'World Best LLM' recruitment, the scale of recruitment, selection procedures, and available budget will be disclosed later.
To respond to the demand for AI computing, the government will secure a total of 18,000 GPUs by the first half of next year. This includes securing 10,000 GPUs by the end of this year for the National AI Computing Center, and 8,000 GPUs to operate the National Supercomputer No. 6 by the first half of next year. For this purpose, 770 billion won will be invested. The government plans to designate AI as a national strategic technology under the Tax Special Treatment Limitation Act to expand tax incentives, and will consider preferential power system impact assessments when constructing AI data centers in non-capital regions, among other improvements to tax and power-related policies.
Finally, to nurture AI talent, the government is pushing to hold a 'Global AI Challenge' and expand AI collaborative research platforms with foreign countries. It will establish programs to support domestic AI researchers and promote the establishment of industry-academia cooperation-type AX (Artificial Intelligence Integration) graduate schools. The goal is to create five unicorn corporations in the AI field by 2027, with policy financing of 5.7 trillion won allocated this year and the establishment of a 3 trillion won scale AI concentration fund by 2027.
The government's establishment of such plans stems from a sense of crisis that it may fall behind in the global AI power competition. Recently, the competition to develop AI models has expanded beyond corporations to a national showdown. Earlier this year, the United States announced the 'Stargate Project,' building a colossal AI infrastructure at a cost of $500 billion (about 730 trillion won). The European Union (EU) also unveiled plans to mobilize a total of 200 billion euros (about 300 trillion won) of private and public capital to promote the AI industry. This dynamic has intensified as China revealed DeepSeek, an AI with performance comparable to global big tech at a low cost earlier this year.
◇ Industry insiders call it 'policy rhetoric lacking innovation'
Currently, global big tech companies are entering the competition to develop artificial general intelligence (AGI), whereas Korea faces a severe shortage of basic infrastructure, such as GPU supply and data center power. As of 2023, Meta and Microsoft (MS) purchased 150,000 units of the NVIDIA high-performance GPU 'H100', while domestic inventory remains around 2,000 units. The industry estimates the total domestic GPU inventory to be at most around 10,000 units.
However, due to the lack of specific budget securing plans being made public, the industry expresses doubts about the effectiveness. A source from the IT industry noted, "The government announced it would introduce 18,000 high-performance GPUs by the first half of next year, which alone requires over 1 trillion won and when you combine installation and other infrastructure expenses, it will likely enter the multi-trillion won scale," and added, "The issues of power problems arising from GPU operation and space allocation are also important, but concrete plans have yet to be announced, raising questions about effectiveness."
There are also criticisms that the scale of support is insufficient compared to that of the U.S. and EU. An IT industry source remarked, "As Korea is already a latecomer in AI, both the government and the private sector must proactively engage in infrastructure building to expect actual effects," stating that "at least 150,000 GPUs should be secured to remain competitive, but the current level is insufficient."
Industry insiders argue that the government's 'World Best LLM' project and AI talent development plans are mere policy rhetoric. An IT industry source stated, "While the government has said it will promote the 'Global AI Challenge' and establish an AI concentration fund to cultivate unicorn companies, this approach is unlikely to secure AI talent."
They added, "AI development is not like the Olympic games," noting that there are more than 10 companies in Korea, including Naver and SKT, that are creating LLMs, and that substantial support should be directed towards these companies. They continued, "To secure AI talent, improving the treatment of domestic AI researchers is a priority, but the core issue isn't being addressed." In fact, compared to last year's starting salaries of 1.2 billion won for Ph.D.-level AI researchers overseas, Korea's is around 400 million won. According to 2022 data, 40% of AI talent who completed their graduate studies in Korea went abroad.
An IT industry source remarked, "While it's fortunate that the government has revealed its support plans, this policy does not reflect the voices of on-site technicians," adding, "AI policy must be approached in the long term, but it is changing at the drop of a hat depending on the regime." They further stated, "Given the current politically unstable situation, there's skepticism about whether this announcement can be realized within the year."
Choi Byeong-ho, a professor at Korea University's Artificial Intelligence Research Institute, said, "In the stage where global AI technology acceleration is taking place, Korea needs three innovative supports—infrastructure, talent, and data—to catch up as a latecomer in AI."