1. Korea's AI Competitiveness: High Rankings, Weak Business

The video opens with the president and Senior Secretary Ha Jung-woo engaged in a candid conversation about Korea's AI competitiveness and its future. The president notes that while Korea ranks 3rd as a group and 10th overall in global AI rankings — higher than its GDP rank — its AI investment and business rankings are not strong.

"When our country is placed in global AI rankings, we're ranked 3rd as a group and 10th overall — higher than our GDP ranking. But apparently our AI investment and business rankings are not high. Why is that?"

Ha Jung-woo identifies weak commercialization and a fragile startup ecosystem as the root causes. He highlights the collision between established industries and AI-native startups, and the lack of institutional and socio-cultural consensus to navigate that transition.

"There are many existing industries — finance, law, education, and more. They all have incumbent players. And then there are companies rising up on the back of AI and other technologies. Naturally, there will be clashes. The question is whether we've built the institutional frameworks and social consensus to let them grow together through a soft landing — or whether they fight it out until one side gets pushed aside. In most cases so far, it's been the latter."

2. The Reality of the Research Environment: GPU Shortages and Administrative Limits

The president expresses dismay at reports of Korean graduate students using PC café GPUs for research. He points out that computing resources are so scarce that students have to change their research topics, and that funding is inadequate.

"A student who had a brilliant topic entered the integrated master's–PhD program at Yonsei University's AI graduate school — only to find there are no GPUs. So they change their topic. They switch to something doable on a small GPU, which means it can't get published in world-class academic journals."

Administrative incompetence is also raised as a concern.

"The problem with the administration is that people without the expertise are still trying to exert control over specialized domains. They don't know, but they try to control anyway."

The president emphasizes that the public sector must shift toward recognizing private-sector expertise and playing a supporting, advisory role.

3. Investment and Institutional Reform for AI Competitiveness

The need for massive investment in computing infrastructure such as GPUs is a recurring theme throughout the conversation. The discussion stresses the importance of concentrated investment to compete with powerhouses like the United States and China.

"In what amounts to an AI arms race, we need to pump substantial resources into research groups and private companies that already have excellent capabilities and can push even further — those who can genuinely compete with the US and China. That's what it will take for us to stay in the game."

The need for autonomy in setting research agendas and flexibility in how research funding is spent is also underscored.

• Scientists should be able to set their own research agendas.
• A portion of research funding should be spendable without receipts.
• Korea's R&D project success rate of 97% is criticized as a structural failure — a sign that researchers only pursue safe, guaranteed outcomes rather than bold, innovative ones.

"Apparently the success rate for our national R&D projects is around 97%. What kind of research has a success rate that high? It means people only pursue what they already know will succeed. No wonder there's no innovation."

4. The DeepSeek Case and the Importance of GPUs

China's AI company DeepSeek is held up as a striking example — its impressive results achieved at seemingly low cost with older GPUs shocked many observers. But the conversation clarifies that DeepSeek actually relied on large quantities of GPUs used at scale.

"DeepSeek's parent company, High-Flyer, already held around 10,000 GPUs by 2020–2021, and now owns approximately 50,000 GPUs including high-performance H100s."

The point is also made that DeepSeek only disclosed the GPU compute costs for training a specific model — the full development cost is far greater.

"The roughly 8 billion won figure cited in the first table about DeepSeek V3 — the AI model known for its strong writing — is not the total development cost. It is only the GPU rental cost for training that one specific model. Personnel costs, data acquisition, and everything else are excluded."

5. The Scale and Operation of GPU Infrastructure

The conversation emphasizes that large-scale GPU concentration and efficient operation are central to winning the AI race. Spreading GPUs across many locations reduces effectiveness; they must be physically co-located and connected at scale.

• Each server holds 8 GPUs; servers are connected via InfiniBand networking.
• Leading global AI companies are now operating clusters of 200,000 GPUs in a single location.
• Korea needs to gain experience operating clusters of 30,000 to 50,000 GPUs.

"The key to operations is to ensure that scale can be fully leveraged — that is what matters. If you buy 30,000 GPUs and distribute 10 to each of 3,000 companies, you're heading down the path to failure."

The need to develop domestic NPUs and GPUs is also raised. Since the United States has begun treating GPUs as strategic assets, Korea must prepare for supply chain risks.

6. Data, Talent, and a Strategy for Global Alliance

Data and talent are identified as other critical pillars of AI competitiveness. While the US and China dominate the technology landscape, a strategy is proposed to partner with nations in the Middle East, Southeast Asia, and Latin America — countries with large volumes of undigitized data.

"Let's form alliances with these countries and jointly undertake projects to digitize the vast analog archives they hold. We have the experience to do this. In return, we get to train extensively on that accumulated data."

Through such global alliances, Korea could lead in multilingual and multicultural AI, potentially positioning itself as a G3 or G4 nation.

7. Defense AI and Rewarding Science and Technology Talent

Given Korea's military density and defense posture, the conversation stresses that drones, robots, unmanned systems, and AI-driven defense are now essential. The low value-added nature of current defense industry exports is criticized, and the need for AI-based defense innovation is argued forcefully.

"I believe our national defense must be transformed by AI. Could we not create new competitions within the defense industry as well? We hear that Korea is becoming an arsenal for the world, but a closer look reveals we're losing enormous core value-added because our manufacturing capabilities are so weak."

The brain drain of science and technology talent, the need for military service exemptions, and the importance of social respect and improved compensation for scientists and engineers are all highlighted.

"After receiving their education, our students go abroad. Why? Simply because the pay is higher. Nations that advance science and technology flourish — and for science and technology to advance, those who work in it must be valued."

"Why does everyone in Korea want to become a doctor? If this continues, won't the country eventually decline? The conclusion has to be: let's build a society that treats scientists with respect and dignity."

8. AI for Everyone in the AI Era

The closing discussion raises the question of how AI will reshape the lives of ordinary citizens. The future of AI, it is argued, hinges not on the technology itself but on the choices society makes.

"What kind of situation AI will put ordinary citizens in is not up to AI. It is up to people."

AI must not become a tool only for the wealthy 1% — it must be made so that everyone can benefit. Special emphasis is placed on the public sector's role in developing AI for vulnerable populations — people with disabilities, the elderly, and children.

"There is something very specific that AI can do — and that is to create technology that helps vulnerable groups, people with disabilities, the elderly, and children actually overcome the difficulties they face in daily life by using AI."

The potential of AI in education to close learning gaps is also highlighted, along with the stated goal of ensuring all citizens have access to generative AI.

"One of my goals is to create opportunities for every citizen to use generative AI."

9. Conclusion: Science, Engineering, and AI Must Thrive for the Nation to Thrive

The dialogue concludes by bringing together all its themes: the importance of science and engineering and AI, combined with national-level investment, institutional reform, talent development, global alliances, and a strong public role — all are necessary for Korea to emerge as a leader in the AI era.

Key Keyword Summary

• AI competitiveness: Strong technological capability, but weak in commercialization and business
• GPU infrastructure: Large-scale concentration and efficient operation are essential
• Research environment: Autonomy, flexibility, and tolerance for innovative failure are needed
• Data acquisition: Multilingual and multicultural data through global alliances
• Talent development: Better compensation, military service exemptions, and social respect
• Defense AI: Transformation through unmanned systems, robotics, and AI-driven defense
• Public sector role: AI for all — especially vulnerable populations and education
• Social consensus: More investment in and respect for science and technology

"The conclusion has to be: let's build a society that treats scientists with respect and dignity."

"What kind of situation AI will put ordinary citizens in is not up to AI. It is up to people."

That covers the key content of the video in chronological order. Science, engineering, and AI must thrive for the nation to thrive! 🚀