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

• Korea ranks 3rd by group, 10th individually in global AI — higher than its GDP ranking. But AI investment and business rankings are not high.
• Root cause: lack of commercialization and startup ecosystem, insufficient regulatory and sociocultural consensus for resolving conflicts between legacy industries and AI-native newcomers.

2. Research Reality: GPU Shortages and Administrative Limits

• Korean grad students research using PC-bang (internet cafe) GPUs, sometimes forced to change topics to fit limited compute.
• Administrative bodies lack technical expertise yet try to control specialized fields.
• The president emphasizes that public institutions should support and take advice rather than control.

3. Investment and Institutional Reform

• Large-scale GPU computing investment is essential to compete with the US and China.
• Research agenda autonomy and spending flexibility are needed.
• Korea's R&D success rate is 97% — "meaning only safe bets are taken. No wonder there's no innovation."

4. The DeepSeek Case

• DeepSeek's parent company Highflyer holds 50,000 GPUs including H100s. The publicized "$80M" development cost was only GPU compute for one training run — not total costs.

5. GPU Infrastructure Scale

• GPUs must be physically co-located in large clusters (30K-50K+ scale), not distributed across thousands of small users.
• Korea needs experience operating at this scale, plus domestic NPU/GPU development to mitigate supply chain risks.

6. Data, Talent, and Global Alliances

• Strategy: Partner with Middle East, Southeast Asia, and Latin American nations to digitize analog data, creating multilingual AI training sets.
• This global alliance can position Korea as a G3/G4 nation.

7. Defense AI and STEM Talent

• Future defense requires drones, robots, unmanned systems, and AI.
• STEM talent exodus: "Students leave because salaries are simply higher abroad."
• "Why does everyone in Korea want to become a doctor? If this continues, the country will eventually decline."

8. AI for Everyone

• "Whether AI pushes ordinary citizens into hardship is not up to AI — it's up to people."
• AI should serve vulnerable populations (disabled, elderly, children) and reduce education inequality.
• Goal: "Give every citizen the opportunity to use generative AI."

"Respect and support scientists — that's the conclusion." "Whether AI pushes ordinary citizens into hardship is not up to AI — it's up to people."