JPMorgan's market intelligence team believes that as AI computing power demand continues to expand, the shipment capacity of AI chips such as NVIDIA GPUs and Google TPUs may become increasingly constrained by HBM and memory supply in the coming years. The material cites a recent statement from Micron stating that even at current memory prices, customer demand significantly exceeds supply, with manufacturers often only able to meet less than half of the demand.
This constraint will also strengthen long-term procurement relationships with large AI companies. SK Hynix, Samsung, and Micron control limited HBM capacity allocations; if customers like NVIDIA and Google prematurely withdraw from long-term procurement agreements, they may face lower supply priority during the next round of capacity shortages.
For AI chip manufacturers, the amount of HBM quota they receive is directly impacting the final delivery volume of GPUs and TPUs.
