피지컬 AI 시대, 반도체 패러다임은 어떻게 바뀌나 | 정인성, 빈센트, 편다송 [밥 보다 투자 하나]

Watch on YouTube ↗  |  January 30, 2026 at 03:08  |  42:19  |  3PRO TV (삼프로TV)
Speakers
Jung In-sung — Author, former SK hynix engineer

Summary

Jung In-sung, a former SK hynix engineer and author, joined Vincent and Pyeon Da-song to discuss how the physical AI era may reshape semiconductors. The conversation focused on whether the memory upcycle can continue, how Samsung and SK hynix are positioned in HBM, Nvidia's robotics ecosystem, TSMC's role in custom HBM, and sovereign AI as a conditional memory demand driver. The main market implication is that memory demand may stay tighter for longer, while company-specific advantages depend on foundry flexibility, IP ecosystems, and cost-performance in robotics.

  • Physical AI and robotics shift the focus toward cost-performance and controlled environments.
  • The DRAM/memory upcycle may not peak in H1 2026 due to long-term contracts and AI demand.
  • Korean memory makers retain a technology and customer-relationship moat against China.
  • Samsung's internal foundry may give it more flexibility in HBM base dies.
  • TSMC's IP ecosystem is an advantage for custom HBM.
  • Nvidia's CUDA, Jetson Thor, and Isaac GR00T stack position it well for physical AI.
  • Sovereign AI could support memory demand but carries overcapacity risk if it fades.
Ideas
Jung In-sung Author, former SK hynix engineer 3:24
Memory upcycle is not peaking yet.
Korean memory makers retain a competitive moat against China: Korea is about four years ahead in DRAM, and US equipment export controls may cap Chinese DRAM at around 10nm-class or fifth-generation levels. As DRAM scaling becomes harder, closer relationships and deeper co-engineering between memory makers and buyers create an additional intangible barrier for incumbents.
000660.KS 005930.KS HIGH
Jung In-sung Author, former SK hynix engineer 5:47
Physical AI robotics growth hinges on cost-performance.
Physical AI and robotics are moving from fixed, controlled factory tasks toward more unstructured environments, where cost-performance becomes the key bottleneck. Cheaper AI, memory, and hardware solutions will determine how quickly robots penetrate homes and other lower-value settings, making physical AI/robotics a major theme to monitor.
ROBO MED
Jung In-sung Author, former SK hynix engineer 10:24
Nvidia dominates physical AI development stack.
In physical AI, raw chip performance and power efficiency matter less than the development environment. Nvidia's CUDA ecosystem, Jetson Thor hardware, and Isaac GR00T software make it easier for robotics developers and partners such as Hyundai's Boston Dynamics and LG to train models and deploy them to robots, so Nvidia is best positioned for robot brains.
NVDA HIGH
Jung In-sung Author, former SK hynix engineer 27:14
TSMC wins custom HBM ecosystem.
As HBM evolves toward custom HBM from HBM4 onward, design IP and the silicon ecosystem become more important. TSMC has a much richer IP and foundry ecosystem, giving TSMC and customers using TSMC silicon more flexibility if custom HBM becomes critical for lowering AI system costs or improving GPU efficiency.
TSM MED
Up Next
30:17
Big Tech Earnings, Elon’s SpaceX–xAI Merge, Genie 3 | Diet TBPN
TBPN · Jan 30
55:21
퇴직연금 수익률 마이너스라고요? 이렇게 하세요! | 김민경, 여도은, 허재무 [아침N투자]
3PRO TV (삼프로TV) · Jan 30
7:08
COMO O BANCO MASTER EXPÔS O LADO MAIS FRÁGIL DO STF?
Market Makers · Jan 30
1:43:19
[체슬리모닝브리프] MS 급락에 혼조 마감...다우 0.11%↑·나스닥 0.72%↓ [26/01/30]
Chesley Investment Advisory (체슬리투자자문) · Jan 30
54:19
Bloomberg Goes One-On-One With NYC Mayor Mamdani | Balance of Power 01/28/2026
Bloomberg Markets · Jan 30

This 3PRO TV (삼프로TV) video, published January 30, 2026, features Jung In-sung discussing 000660.KS, 005930.KS, ROBO, NVDA, TSM. 4 trade ideas extracted by AI with direction and confidence scoring.

Speakers: Jung In-sung  · Tickers: 000660.KS, 005930.KS, ROBO, NVDA, TSM