AI speed-control controversy, will infrastructure investment also waver? What we ultimately need to watch is 'this' | Kim Jang-yeol, Center Head at Unistory Asset Management

AI speed control controversy, will infrastructure investment also waver? What we ultimately need to watch is 'this' | Kim Jang-yeol, Center Head at Unistory Asset Management [Global Interview]
Watch on YouTube ↗  |  September 13, 2026 at 22:43  |  31:50  |  3PRO TV (삼프로TV)
Speakers
Kim Jang-yeol — Reporter, The Bell

Summary

Kim Jang-yeol, Center Head at Unistory Asset Management, discussed whether the renewed AI safety and speed-control debate should change AI infrastructure and semiconductor investment. He argued the controversy mainly hurts sentiment and is not yet evidence of a data-center capex slowdown, with data-center spending and big-tech comments the key monitors. He also covered Fed rate-hike expectations, the higher-for-longer rate regime, Korean exporter FX headwinds, SK hynix valuation, long-bond duration potential, and Apple's on-device AI/NAND implications.

  • AI safety debate dominated weekend sentiment; speaker said it is too early to equate it with an infrastructure capex slowdown.
  • Data-center company capex is the key indicator to watch, especially around late-October earnings and commentary.
  • September Fed hike is largely priced; higher-for-longer rates are becoming the new normal.
  • Long-term bonds could eventually rally if AI-led growth supports the economy after the hiking cycle.
  • Stronger won is expected to lower Korean exporter earnings, with autos more exposed.
  • SK hynix was described as Micron-level in fundamentals with a still-low mid-6x P/E and the worst likely over.
  • Apple's foldable and on-device AI strategy is not viewed as negative, but product reception and memory mix remain uncertain.
  • On-device AI could favor NAND over DRAM because of model compression and modular storage needs.
Ideas
Kim Jang-yeol Reporter, The Bell 2:48
AI infrastructure capex has not broken.
The renewed AI safety and speed-control debate is a negative sentiment event, but it is too early to conclude that it will slow AI infrastructure, GPU, or data-center investment. Safety reinforcement may require more investment rather than less, and the decisive evidence will be whether major data-center companies and big tech keep raising capex; their late-October earnings and commentary are the key checkpoint.
Kim Jang-yeol Reporter, The Bell 20:31
AI growth may lift long bonds.
The Fed's September hike is largely priced, and the new normal is a structurally higher rate level rather than a quick return to ultra-low rates. If AI-led growth keeps the economy resilient after the hiking cycle, long-term yields could eventually fall as investors buy duration, lifting long-term bond prices.
Kim Jang-yeol Reporter, The Bell 21:54
SK hynix cheap, worst over.
SK hynix's fundamentals are now roughly Micron-level, and its historical 20-50% valuation discount to Micron has largely closed in the ADR. Even assuming next-year earnings are cut by more than 10% on FX, it trades around a mid-6x P/E, so overall valuation remains low and the worst may be over; the remaining issue is the still-wide ADR/local-share gap.
Kim Jang-yeol Reporter, The Bell 24:59
Strong won hurts Korean exporters.
The won has already strengthened significantly, and because most Korean listed companies are exporters, FX translation should lower Q3/Q4 and next-year earnings forecasts. Samsung Electronics and SK hynix are exposed, shipbuilders less so, and automakers are described as even more exposed.
Kim Jang-yeol Reporter, The Bell 27:12
Apple AI phone needs proof.
Apple's foldable iPhone is late but looks like a strategic response to surging memory costs rather than a negative signal, and the more important question is whether Apple can implement on-device AI in a realistic modular way. Until the product is in users' hands and consumer response is visible, this is a wait-and-see catalyst.
Kim Jang-yeol Reporter, The Bell 29:12
On-device AI favors NAND.
Apple's on-device AI approach is not to run PC-like AI on the phone but to keep a basic model and attach specialized modules as needed. That model-compression and modular architecture may mean DRAM is used less, while greater local storage needs make NAND relatively more attractive.
Up Next

This 3PRO TV (삼프로TV) video, published September 13, 2026, features Kim Jang-yeol discussing AIQ, TLT, 000660.KS, Korean exporters, Korean automakers, AAPL, NAND flash memory. 6 trade ideas extracted by AI with direction and confidence scoring.

Speakers: Kim Jang-yeol  · Tickers: AIQ, TLT, 000660.KS, Korean exporters, Korean automakers, AAPL, NAND flash memory