NVIDIA Earnings D-1, 'This' Is More Important Than the Numbers... AI Semiconductor Demand Hasn't Broken Yet | Kim Jang-yeol, Head of Research Center, Unistory Asset Management
NVIDIA Earnings D-1, 'This' is More Important Than the Numbers... AI Semiconductor Demand Hasn't Broken Yet | Kim Jang-yeol, Head of Research Center, Unistory Asset Management [Global Interview]
Watch on YouTube ↗
|
August 25, 2026 at 22:39
|
31:21
|
3PRO TV (삼프로TV)
Ideas
Nvidia's AI financing narrative matters most.
Nvidia's recent seven-session drop is small and not meaningful. Over the past three years, Nvidia usually rallied into earnings and sold off after results even when earnings beat; actual reported numbers have been strong with revenue about 6% and EPS about 10% above consensus, but guidance matters. For long-term investors, weakness after the event is a buying opportunity while holders should hold.
Token growth keeps AI semis demand intact.
AI semiconductor demand has not broken because total token volume is the real driver. Google's token usage grew roughly 300x over two years, about 15-20x per year. Token prices fall 80-90% annually, so total token volume needs to grow more than 10x per year for revenue and capex to keep working. Agentic AI, with multiple agents and many more iterations, supports this token growth outlook.
Oracle is AI infrastructure weak link.
Oracle is identified as the relatively weak link in the AI infrastructure chain due to its OpenAI-related exposure; it represents about 25% of AI demand. If Oracle struggles, the remaining 75% can still support AI demand, but Oracle itself carries company-specific risk.
Nebius rebounds on backlog and financing.
The neocloud group, such as Nebius, was previously seen as the weak AI funding link, but Nvidia's willingness to lend into AI infrastructure and their own backlog of about three times capex changed that perception. Their stocks rebounded in July-August as the market recognized that funding and demand were available.
Energy scarcity attracts fund buying.
Both hedge funds and mutual funds have been buying energy because AI-driven power demand makes energy scarce; they have also bought nuclear/power generation names. This is a common fund-flow signal that energy is a scarce input to the AI buildout.
Mutual funds are turning to memory stocks.
US mutual funds started accumulating memory names about one to two weeks ago, including SanDisk, Micron, and SK hynix ADR, while hedge funds had sold memory from June. Korean memory weakness was driven more by hedge fund supply and local flow issues than by memory fundamentals; the turn in fund flows supports the memory complex.
Marvell custom AI ASIC networking play.
Marvell is another AI semiconductor axis beyond Nvidia: it supplies custom ASIC-related networking chips for Google TPU and Amazon Trainium, and foreign banks raised target prices. Its improving sentiment signals both Nvidia and custom ASIC demand are healthy, supporting the broader semiconductor demand picture.
Semiconductor worst is likely over.
The negative AI/semiconductor sentiment appears washed out. Kim says the worst is over and a new low is unlikely unless several shocks occur simultaneously — two more Fed rate hikes, oil spiking to $110, and Nvidia/Marvell results falling apart — because underlying AI demand and semiconductor fundamentals remain intact.
This 3PRO TV (삼프로TV) video, published August 25, 2026,
features Kim Jang-yeol
discussing NVDA, SMH, ORCL, NBIS, XLE, URA, MU, SNDK, SK Hynix ADR, MRVL, SOC.
8 trade ideas extracted by AI with direction and confidence scoring.
Speakers:
Kim Jang-yeol
· Tickers:
NVDA,
SMH,
ORCL,
NBIS,
XLE,
URA,
MU,
SNDK,
SK Hynix ADR,
MRVL,
SOC
