Nvidia CEO Jensen Huang has dismissed predictions that the memory chip market will tip into oversupply by 2029, telling CNBC at the company’s GTC Berlin conference that AI infrastructure demand will keep absorbing every chip the industry can make. He also left the door open to working with Chinese memory makers and said US export restrictions have only accelerated China’s drive for chip self-reliance.
“The More You Have, The Better”
Asked in an on-site interview on October 22 whether he feared a memory supply glut by 2029, Huang replied, “I don’t think so.” He said, “One of the advantages of memory is that you can always use more,” and predicted that demand for memory would keep rising as AI spreads through the economy.
He singled out high-bandwidth memory, the stacked DRAM that sits next to AI accelerators, saying “HBM will continue to be used for years to come.” He also said Nvidia would keep using large quantities of LPDDR, the low-power memory that originated in mobile phones and which Nvidia was the first to bring into data centres — a move he said delivered twice the energy efficiency of anything else on the market.
An Answer To The Memory Sell-Off
Huang’s comments were aimed directly at a recent wave of pessimism about memory stocks. South Korean press forecasts have suggested the semiconductor supercycle could end in 2029 as expanded memory supply catches up with demand, and China’s ChangXin Memory Technologies (CXMT) announced next-generation DRAM design technology that investors read as a threat to the incumbents.
The result was a sharp sell-off in memory-related stocks, including Samsung Electronics, SK hynix and Micron — the three suppliers that dominate the HBM market Nvidia depends on. Huang’s argument, in essence, is that AI training and inference workloads will consume the new supply: the glut scenario assumes demand plateaus, and he does not believe it will.
Open To Chinese Suppliers, Frank About Export Controls
Asked whether Nvidia would consider using CXMT products, Huang said, “We have an open mind,” adding that “corporations around the world are working with Chinese companies at some point in their supply chains.” He praised China’s manufacturing capability — “In manufacturing, there is no country superior to China when looking at the nation as a whole” — and said US restrictions on semiconductor exports to China have “if anything, accelerated China’s technological self-reliance.”
The remarks land days after Huang dismissed AI-extinction warnings as “cheap” headline-grabbing at the same Berlin conference. Both comments fit the same posture: the AI buildout is real, demand is real, and obstacles — whether safety panic or supply pessimism — are overblown.
“Demand For Neocloud Is Beyond Imagination”
Huang also addressed the broader financing anxiety hanging over AI infrastructure. After declines in cloud-infrastructure stocks such as CoreWeave, he said, “Demand for neocloud is beyond imagination” — what constrains the industry’s growth rate is not a lack of demand but a lack of working capital. “Across the AI industry, demand for capital far exceeds available capital,” he added, “This issue will eventually be resolved.”
That echoes his push to make Nvidia’s compute a “financeable asset” — hardware that lenders and investors can underwrite — as the company tries to keep the data-centre buildout funded. For memory makers, his message was simpler: keep expanding, because Nvidia will keep buying.
Why It Matters
Memory is the quiet constraint of the AI boom. Every GPU Nvidia ships carries HBM from Samsung, SK hynix or Micron, and the memory supply chain is already being stretched — Samsung is tapping TSMC’s 3nm process for Nvidia’s next-generation NVHBM memory. A 2029 glut would collapse HBM prices and margins for the memory makers; a sustained shortage keeps prices high and could throttle AI hardware production.
Huang’s intervention matters because he is the demand side speaking to the supply side. As the buyer of the industry’s HBM output, his confidence that “you can always use more” is the strongest signal memory investors could get — and his openness to Chinese suppliers is a reminder that supply can come from unexpected places if the incumbents overbuild.
Background
Nvidia has spent October deepening its hardware partnerships: the $4,999 DGX Spark 64GB reached partners this month, and Amkor signed a $1.3 billion deal to expand AI chip packaging in South Korea. Huang’s GTC Berlin keynote has become a running commentary on the economics of the AI buildout, following his dismissal of AI-extinction warnings at the same event earlier in the week.
Source: Chosun Biz ; Asia Business Daily
0 Comments