OpenAI 'Jalapeno' Appears… Is Nvidia's Solo Run Over? / The Golden Time to Check China Is Over... Samsung Electronics and SK Hynix's Real Crisis Begins? | Industry Analyst Dr. Lee Ju-wan

OpenAI 'Jalapeno' Appears… Is Nvidia's Solo Run Over? / Golden Time to Check China is Over.. Samsung Electronics and SK Hynix's Real Crisis Begins? | Industry Analyst Dr. Lee Ju-wan
Watch on YouTube ↗  |  August 29, 2026 at 06:00  |  23:02  |  815 Money Talk (815머니톡)
Speakers
Lee Ju-wan — Industry Analyst
Kim Tae-seong — Division Head, Xangle

Summary

Industry analyst Lee Ju-wan discusses OpenAI's custom inference chip as part of a structural shift from Nvidia's GPU dominance toward ASIC and customized AI silicon. He argues Nvidia's dependency will decline over time even if its near-term leadership remains. For Korean memory, he sees multiple risks: Samsung and SK hynix are ceding share by cutting production, Chinese memory players are becoming self-sustaining, the stronger won will compress margins, and US antitrust litigation creates a large but uncertain legal overhang. Micron is framed as a relative beneficiary of Korean production discipline.

  • OpenAI's custom inference chip signals the broadening inference market and shift toward customized AI silicon.
  • ASIC share versus GPU is shifting from roughly 9:1 to 7:3, with Broadcom and Marvell as key players.
  • Samsung and SK hynix cut NAND output to support prices, ceding share to China and Micron.
  • YMTC has crossed the 10% NAND share survival line; CXMT DRAM may cross 10% by year-end.
  • Chinese memory cost advantages and self-funded capacity end the Korean-led oligopoly period.
  • A stronger won plus memory price declines may hit Samsung and SK hynix earnings harder than Micron or Chinese peers.
  • US court antitrust review of memory output restraint creates a large but uncertain legal risk.
Ideas
Lee Ju-wan Industry Analyst 3:37
Nvidia dependency is structurally declining.
OpenAI's custom inference chip 'Jalapeno' is an early sign that inference demand is shifting away from Nvidia's standardized GPU stack. Nvidia remains the near-term leader because of its installed base and volume, but customers increasingly want lighter, cheaper, lower-power silicon customized to specific inference tasks. Nvidia dependency will therefore structurally decline, and even if Nvidia keeps first place, its share may erode from about 90% toward 70% or 60% over time, which can eventually affect the stock.
Lee Ju-wan Industry Analyst 4:35
ASIC custom AI silicon is taking share.
The main expression of reduced Nvidia dependence is the rise of ASIC and custom AI semiconductors. GPUs improve performance mainly by adding more dies, making them increasingly heavy, expensive and power-hungry, while inference customers want lighter, application-specific chips using more CPU/NPU and less GPU. The ASIC share of AI silicon has already moved from about 9:1 to 7:3 versus GPUs and is likely to keep rising, with Broadcom as the incumbent and Marvell Technology emerging as a competitor.
Lee Ju-wan Industry Analyst 8:22
Korean memory cuts are ceding share.
The intentional underutilization that supported memory prices has now been exposed, and a US California federal court began an antitrust/price-fixing review on June 25. The eventual damages are unforecastable: they could range from tens of trillions of won to more than one quadrillion won across the industry. This creates a large binary legal overhang for Samsung Electronics, SK hynix and Micron, and the speaker argues buybacks and dividends should be deprioritized in favor of building loss provisions.
Lee Ju-wan Industry Analyst 8:49
Micron benefits from Korean production cuts.
The intentional underutilization that supported memory prices has now been exposed, and a US California federal court began an antitrust/price-fixing review on June 25. The eventual damages are unforecastable: they could range from tens of trillions of won to more than one quadrillion won across the industry. This creates a large binary legal overhang for Samsung Electronics, SK hynix and Micron, and the speaker argues buybacks and dividends should be deprioritized in favor of building loss provisions.
Up Next

This 815 Money Talk (815머니톡) video, published August 29, 2026, features Lee Ju-wan discussing NVDA, AVGO, MRVL, SMH, 005930.KS, 000660.KS, MU. 4 trade ideas extracted by AI with direction and confidence scoring.

Speakers: Lee Ju-wan  · Tickers: NVDA, AVGO, MRVL, SMH, 005930.KS, 000660.KS, MU