JP Morgan's markets intelligence team says that as AI compute demand expands
over the next few years, shipment capacity for Nvidia GPUs, Google TPUs and
similar AI chips could increasingly be constrained by HBM and memory supply.
Micron recently said customer demand remains materially higher than supply even
at current memory prices, and vendors often meet less than half of demand. SK
Hynix, Samsung and Micron control limited HBM capacity, reinforcing large
buyers' long-term procurement ties; buyers who exit long-term contracts may face
lower allocation priority in the next capacity crunch. HBM allocations are
directly limiting how many GPUs and TPUs can be delivered.
