Jensen Huang: No Risk of Memory Oversupply Before 2029
nashnova research
Nvidia CEO Jensen Huang told CNBC that memory oversupply before 2029 is highly unlikely, directly countering the fear that hammered storage stocks over the past week — a statement aimed squarely at reassuring investors in SK Hynix, Samsung, and Micron.
Why did memory stocks suddenly sell off?
On October 15, a prominent South Korean outlet published commentary citing two experts who predicted a "new flood of supply" in the memory market by 2029 — turning it into a "red ocean" and potentially ending the AI super-cycle.
The article triggered concentrated selling in SK Hynix, Samsung, and Micron; Elon Musk publicly echoed the concern, saying China "will mass-produce memory by 2028."
This means → the market's deepest fear isn't today's demand — it's a supply surge three to four years out that crashes prices back to commodity levels.
What's Huang's case for "no oversupply"?
Huang offered a foundational argument: "Memory has a nice characteristic — you can always use more of it."
In plain terms = memory isn't like oil, where consumption has a ceiling. As AI models grow larger and inference scales up, memory consumption expands like a sponge — the demand ceiling keeps rising.
He also revealed that Nvidia has pioneered bringing LPDDR memory — the low-power type used in smartphones — into data centers, achieving twice the industry's best energy efficiency.
HBM and LPDDR — what do two memory tracks mean?
Huang stated clearly that HBM (High Bandwidth Memory) will remain widely used for years — it is the core memory solution for today's AI chips.
LPDDR entering the data center is a new move. In plain terms = smartphone-grade power-saving memory running server workloads — same output, half the electricity.
This reflects a strategy beyond "buy more memory" — Nvidia is blending different memory types to drive down total power consumption.
Could China's CXMT steal market share?
CXMT (ChangXin Memory Technologies) recently unveiled a new architecture, raising concerns it could take share from SK Hynix, Samsung, and Micron.
Huang's response was surprisingly open — he said Nvidia is open to working with CXMT: "We work with Chinese companies; virtually every company in the world works with Chinese companies somewhere in the supply chain."
He also praised China's manufacturing capability: "No country exceeds China in overall manufacturing," adding that China's technical strength must be taken seriously.
Chip export controls — containment or catalyst?
Huang said plainly that U.S. chip export restrictions have objectively accelerated China's self-developed capabilities. "China's pace has certainly quickened because of these regulations," he said.
In plain terms = the intent was to bottleneck a rival; the result was to force that rival to run faster — controls became a catalyst.
He added that U.S. tech companies remain strong and can handle competition from any direction.
Can one statement move the market — and what comes next?
Whether Huang's remarks can truly reverse market expectations remains to be proven.
The key variables are SK Hynix, Samsung, and Micron's upcoming capex plans and capacity data — only if actual capacity growth stays below demand growth can "no oversupply" hold.
This means → the direction for memory stocks won't be set by who talks loudest, but by where the capital actually flows.
市场有风险，内容仅供研究参考，不构成投资建议。