South Korea Boosts AI Infrastructure with 10,000 New GPUs and National LLM Initiative
The South Korean government is set to significantly enhance its artificial intelligence (AI) capabilities with the upcoming distribution of 10,000 high-performance graphics processing units (GPUs). These vital components, acquired through a supplementary budget from the previous year, are slated for allocation as early as March. The announcement was made by Ryu Je-myung, the Second Vice Minister of the Ministry of Science and ICT, during a recent work report session with various affiliated institutions and public agencies.
Vice Minister Ryu detailed the government’s strategic approach to distributing this crucial hardware. “We plan to supply 10,000 of the latest GPUs secured through last year’s supplementary budget this year,” he stated, emphasizing that a comprehensive demand survey for GPUs was conducted up until December. To ensure a fair and effective allocation, the government will establish an evaluation committee by February. This committee will meticulously assess incoming applications, paving the way for the commencement of GPU supply in March.
The allocation strategy prioritizes national strategic interests and the advancement of the AI ecosystem. Vice Minister Ryu outlined the planned distribution percentages:
* 50% of the total GPU quantity will be dedicated to national projects, underscoring the government’s commitment to leveraging AI for major national initiatives.
* 30% will be directed towards the industrial sector, aiming to foster innovation and competitiveness within South Korean businesses.
* 20% will be reserved for academia and research institutions, supporting cutting-edge research and the development of future AI talent.
A key consideration in the GPU supply timeline is the upcoming operational launch of the Korea Institute of Science and Technology Information (KISTI)’s sixth supercomputer in July. “Since KISTI will begin operating the sixth supercomputer in July, we plan to focus on supplying the GPUs before then,” Vice Minister Ryu added, highlighting the urgency to bolster the nation’s supercomputing power.
Beyond hardware, South Korea is also making strides in developing its own foundational AI technologies. Minister Bae Kyung-hoon of the Ministry of Science and ICT has put forth a proposal to advance KISTI’s science and technology-specialized large language model (LLM), named ‘KONI’, into Korea’s independent AI foundation model.
KONI, first introduced by KISTI in late 2023, is designed to support research within government-funded institutions. Its core functionality lies in its ability to learn from the vast amounts of science and technology big data generated through national research and development efforts. This specialized training allows KONI to understand and process complex scientific information, making it a powerful tool for researchers.
Minister Bae expressed optimism about the potential of KONI. “The results of the independent AI foundation model have already been released as open source, and their evaluation results are also available,” he remarked. While acknowledging that further training will be necessary, he conveyed the government’s aspiration to “proceed with updating KONI as our Korean model this year.” This initiative signifies a strategic move towards achieving AI sovereignty and reducing reliance on foreign-developed foundation models. The open-source release of its initial results indicates a commitment to transparency and collaborative development within the AI community. The ongoing efforts to refine and expand KONI’s capabilities are expected to yield significant advancements in South Korea’s AI research and development landscape.