In 2026, South Korea's artificial intelligence (AI) budget is set at 9.9 trillion won, three times that of the previous year. In August, construction began on the National AI Computing Center in Haenam, and the 'Artificial Intelligence Basic Law' took effect in January. Additionally, a project to develop a unique foundation model is underway. These figures indicate an unprecedented scale. However, in the same year, the four major U.S. tech companies announced capital investment plans totaling $725 billion, approximately 100 times South Korea's budget. Just Amazon's quarterly spending exceeds the entire annual AI budget of South Korea. The investment is projected to triple from $226 billion in 2024 within two years, highlighting a gap that cannot be closed through similar strategies.
The United States and China are advancing in different ways. The U.S. relies on autonomous competition from private capital, while China allocates resources through long-term national planning. Neither model is one that South Korea can simply replicate. Competing with capital against the U.S. is unfeasible, and matching China's scale through state mobilization is also unrealistic. However, China's ability to innovate under export controls, achieving similar performance with less computational power, is noteworthy as it demonstrates how constraints can spur innovation.
President Lee Jae-myung has consistently emphasized the goal of becoming one of the top three AI nations since his election campaign. In February, the 'AI Basic Plan' was released, and the budget of 9.9 trillion won was allocated. However, a clear roadmap outlining who should do what and by when is still lacking. While the goal is clear, the path remains vague. As a scholar who has researched and taught AI for 42 years, I propose a roadmap detailing what five key stakeholders—government, local governments, universities, businesses, and individuals—should do by 2030.
Returning to the earlier figures, the question we should ask is not 'How do we catch up?' but 'What do we define as third place?' This is not a mere play on words; how we define our goals significantly alters the actions we take and our chances of success.
The top three AI nations can be defined in three categories. The first is 'Model Power,' which involves possessing world-class foundation models. Frankly, this is challenging as the required capital exceeds what we can mobilize. The second is 'Utilization Power,' which focuses on leading AI transformation across industries and the public sector, a more achievable goal. South Korea is a society that quickly adopts new technologies, and there are numerous opportunities for application. The third is 'Infrastructure Power,' which pertains to our status in the AI supply chain, including semiconductors, data centers, and power, an area where we have already made significant progress.
Thus, a clear order emerges. We should leverage our infrastructure to enhance utilization, and through the data and capital accumulated in this process, target specific areas for model development. Pursuing all three simultaneously may scatter our resources, but establishing a sequence can create synergies. Let's assess what we have. High Bandwidth Memory (HBM) is an essential component of AI accelerators, and South Korean companies lead this market. Few countries have a structure where increasing investments in AI data centers yield greater profits. It is also rare to possess shipbuilding, steel, petrochemicals, automotive, and battery manufacturing capabilities all at once.
In the realm of physical AI, where AI extends beyond software to directly handle robotics and equipment, countries with actual manufacturing experience hold an advantage, as process data cannot simply be scraped from the internet.
However, there are clear weaknesses. According to a report from Stanford University, South Korea ranks fourth overall and first in AI patents per capita. In contrast, the International Institute for Management Development (IMD) ranks South Korea 15th in digital competitiveness, a drop of nine places, with the talent sector at 49th and foreign talent inflow at the bottom tier. The discrepancies in these indicators arise from differing measurement criteria. While our manufacturing capabilities are strong, our human resources and market conditions are average. This asymmetry accurately reflects the state of AI in South Korea.
More concerning is that lower rankings can undermine higher ones. If talent continues to leave, research outcomes will decline in a few years, and a thin industrial ecosystem will prevent good technologies from translating into business success. Our current high rankings are sustained by past achievements. Another critical point is energy. Surprisingly, the last bottleneck in AI competition is electricity. A large data center consumes as much power as a small city, yet demand is concentrated in the metropolitan area while power generation facilities are dispersed in rural areas. The decision to locate the National AI Computing Center in Haenam addresses this structural issue. However, the challenges related to grid connection schedules, permitting speed, cooling water, and community acceptance are not purely technical; they are administrative and consensus-based issues.
It is essential to emphasize that a nation's AI capabilities are determined not by the top 1% but by what the remaining 99% can achieve. While securing top researchers is crucial, it alone will not elevate us to the top three. With the rise of vibe coding, where AI generates code based on natural language descriptions, the bottleneck is shifting. We are moving from a focus on the number of developers to the number of ideas. Previously, many good ideas disappeared due to a lack of implementers.
Now, those who understand the problems best can create tools directly. A factory supervisor can develop a monitoring program tailored to their process, a nurse can create a handover organization tool, and a civil servant can design a complaint classification tool. Such tools cannot be outsourced, as the problem-solver and the creator are often different individuals.
Of course, there are risks. Using code without understanding its function can lead to security vulnerabilities or incorrect calculations entering the field. Therefore, training in both tool creation and critical evaluation of results must go hand in hand. However, this observation does not imply 'do not teach' but rather 'teach properly.' To become an AI powerhouse, we must first become a 'vibe coding powerhouse.' This strategy cannot be realized by the government alone.
The government must lay the groundwork by providing computational resources, data, and electricity while establishing predictable regulations. Local governments should not stop at attracting data centers but should connect this to the transformation of regional key industries. If the allocation ratio of computational resources for local businesses and universities is not specified in the attraction contracts, only energy-consuming buildings will remain in the region. Universities should integrate AI into all majors rather than merely increasing the number of AI majors. It is faster to teach AI to domain experts than to teach domain knowledge to AI specialists. Businesses must distinguish between adopting tools and redesigning work processes. Even if a report draft is created in five minutes, if the approval process remains unchanged, the overall time will not decrease. Individuals should break down their jobs into task units before contemplating a career change, as it is the tasks within jobs that may disappear, not the jobs themselves.
If any one of the five stakeholders falters, the efforts of the others will not yield results. If the government secures GPUs but lacks researchers to utilize them, they will remain idle. If universities train talent but there are no companies to employ them, they will seek opportunities abroad. If businesses adopt AI but employees do not use it, costs will only accumulate. Ultimately, national strategies must translate into someone's Monday morning tasks to be realized. What is lacking is not the budget but the decisions. Defining the scope of AI learning from copyrighted materials, establishing performance indicators and termination criteria for all projects, and shifting the recruitment of foreign talent from salary negotiations to package design do not require vast sums of money. The longer the regulatory gaps persist, the more disadvantaged domestic companies become.
There are four budget cycles remaining until 2030. This is not a long time. However, if the direction is correct, it can also be sufficient. We must be cautious that if rankings become the goal, we will only pursue what can be measured. If we can solve the care needs of regions with declining populations, address workforce shortages in fields unable to find personnel, and preserve vanishing skilled technologies through this technology, then achieving third place will be a result, not a goal.
Author's Major Background
▷ Ph.D. in Business Administration from Sungkyunkwan University ▷ Senior Researcher at the East-West Economic Research Institute ▷ Reporter at Maeil Business Newspaper ▷ President of the Korean Management Association ▷ President of the Sustainable Science Society ▷ Dean of the Global Convergence College at Sookmyung Women's University, Dean of the Hallyu International College, Professor of the Convergence International Department and Hallyu International Department ▷ President of the Korean AI Education Association ▓ President of the Korean Vibe Coding Association ▓ Director of the Moon Hyung-nam Branding Research Institute
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.