Huang Dismisses 2029 Memory Glut Fears as NVIDIA Touts Memory Efficiency

NVIDIA CEO Jensen Huang pushed back against investor concern about a potential memory market oversupply by 2029, saying data centers can always absorb more memory. The concern was triggered by a South Korean opinion column published October 15 warning of a supply flood turning the market into a "red ocean" by 2029, with two experts cited in the column predicting the AI memory supercycle could end and the industry could enter a downturn by then. Memory stocks fell further after China's CXMT announced a new design architecture that raised investor fears it could take market share from SK Hynix, Samsung, and Micron. Separately, NVIDIA has described a full-stack codesign approach it says improves memory utilization across its hardware and software stack, citing techniques including NVFP4 precision, KV cache offloading via Dynamo software, and distributing model weights across up to 72 GPUs in an NVL72 system.