[To the New President] 100 Trillion Won for AI Investment? "LLM Development Is Good, but Efficiency Must Improve"

President Lee Jaemyung's pledge to invest 100 trillion won in the development of the artificial intelligence (AI) industry has drawn criticism from the AI sector, which warns that unless such a massive budget and resources are allocated appropriately, it could result in a waste of taxpayer money. One example cited is the domestic support program for inference AI semiconductors, specifically NPUs (Neural Processing Units). Just before the presidential election, President Lee pledged to "support the development of next-generation semiconductor technologies such as low-power, high-performance NPUs and PIMs, and to establish an AI semiconductor ecosystem centered on domestically produced AI semiconductors at an early stage."

However, in reality, there are not many companies in the field that actually require NPUs. According to the "2024 AI Industry Status Survey Report" by the Software Policy & Research Institute, which is affiliated with the Ministry of Science and ICT, 7 out of 10 domestic AI companies responded that they have neither plans nor intentions to adopt NPUs. Only 23.8% said they were considering adopting NPUs. In this context, Cho Hyungkeun, Chief Strategy Officer (CSO) of Moreh, a company specializing in AI computing infrastructure software, commented, "Before supporting the production of AI semiconductors, it is more important to first secure system software (SW) capabilities that bridge AI semiconductors and applications. Under the current circumstances, even if NPUs are produced, there are not many companies that will use them." He advised, "If the new government wants to foster the AI industry, it must actively reflect the voices from the field."

President Lee has also pledged to secure more than 50,000 GPUs (Graphics Processing Units) to establish a national AI data cluster and to develop a national large language model (LLM). He stated that these initiatives would create an AI innovation ecosystem and nurture AI as a future strategic industry. The AI industry has responded to these pledges with both anticipation and concern.

Hwang Rigeon, co-founder and Chief Platform Officer of WantedLab, an AI-based talent recruitment platform, said, "If domestic AI companies are to compete with global big tech firms, differentiated strategies and practical approaches are crucial. They must develop AI solutions that deliver real value to users." He added, "Rather than focusing on LLM competition, a more realistic alternative is to create AI services that address user inconveniences and boost productivity, thereby targeting the global market."

Kim Juwon, CEO of Coxwave, which operates the AI product analytics platform Align, stated, "The core of AI industry competitiveness lies in effectively utilizing the latest technologies to deliver exactly the features users want, and in enabling tasks that previously required hundreds of people to be accomplished by just a few. This is where the new government's resources should be invested." As examples, he cited DeepSeek, which has released low-cost, high-quality AI models, and Perplexity, which is emerging as a rival to Google and has launched the high-performance AI model 'Sonar' based on Meta's Llama LLM.

Industry leaders also unanimously called for reducing uncertainty regarding the AI Basic Act, which will take effect in January next year. Hwang expressed concern, saying, "The definition of high-impact AI is so abstract and broad that the law could function as excessive regulation. There is a risk that the measures could be applied more strictly to domestic companies than to foreign ones, thereby hindering domestic technological development and creating reverse discrimination." Kim also stressed the need for specific guidelines, stating, "Regarding the provisions on AI reliability and safety stipulated in Articles 32 and 34 of the AI Basic Act, we need clear instructions on how to ensure reliability and safety, and what safeguards companies must establish to meet these requirements."