Nvidia chief executive Jensen Huang said he does not expect an oversupply of memory semiconductors before 2029, pushing back on fears that a wave of new capacity could swamp the market, according to KB Think.
Asked directly about the risk of a memory glut, Huang dismissed the idea, saying the defining trait of memory is that having more of it is always better because systems can always consume additional capacity. He added that demand for memory will keep climbing.
The remarks, reported by KB Think and carried by several Korean outlets including SBS News, KBS News, News1 and Hankyung, come as investors weigh whether heavy spending on memory production could eventually outrun demand.
Why it matters
Memory pricing and availability sit at the center of the artificial intelligence buildout. Training and running large models depends on high-bandwidth memory (HBM) stacked next to Nvidia's accelerators, and any shift from shortage to surplus would ripple through chipmakers, server builders and cloud operators.
Huang's comments are a signal to suppliers that Nvidia expects to keep absorbing large volumes of memory for years rather than bracing for a downturn. They also matter for Korean memory makers, whose products have been closely tied to Nvidia's roadmaps, and for Chinese suppliers hoping to break into the same supply chains.
HBM and LPDDR stay in the plan
Huang said HBM, the high-bandwidth memory used alongside AI processors, will remain in use for years to come. He also confirmed that Nvidia intends to keep using LPDDR, a low-power memory type, in large quantities.
According to KB Think, Huang described Nvidia as the first company to bring phone-grade LPDDR memory into data centers worldwide. He said this approach delivers roughly twice the energy efficiency of any product available today.
That claim places power consumption at the heart of Nvidia's pitch. Data centers running AI workloads are constrained by electricity and cooling as much as by chip supply, so memory that draws less power per unit of work is a selling point for operators expanding capacity.
Open door to CXMT
Huang was also asked whether Nvidia would use products from ChangXin Memory Technologies (CXMT), the Chinese memory maker that has been working to catch up with established Korean suppliers. He said the company keeps an open mind.
He went on to praise China's capabilities, saying the country is remarkably strong in foundational infrastructure technology and that no other nation matches it across manufacturing at a national scale. He said those capabilities deserve to be taken very seriously.
The answer does not commit Nvidia to any purchase, but it signals that the company is not restricting itself to a single country's suppliers. For CXMT, a favorable word from the head of the world's most valuable chip company carries weight even without a contract attached.
Backing the neoclouds
Huang also addressed concerns around so-called neocloud companies — specialist AI cloud providers such as CoreWeave, Nebius and IREN — that have drawn scrutiny from investors. He said demand for their services is exploding beyond what charts can capture.
According to KB Think, Huang argued that what limits these companies is not demand but working capital. He said the value they provide is what ultimately matters, and that the funding question will be resolved on that basis.
That defense matters because neoclouds are among the largest buyers of Nvidia hardware, and their ability to keep raising capital has become a focus for investors watching the AI trade. Huang's framing casts their constraints as a financing issue rather than a sign of weakening appetite for compute.
What to watch
The next signals will come from memory makers' capacity plans and from Nvidia's own purchasing decisions, including whether any Chinese memory products appear in its supply chain.
