GPU computing power has been pushed to the limit, but it's bottlenecked by memory bandwidth — Jukan's view is quite counterintuitive, but upon reflection, HBM's status in the field needs to be re-evaluated.

CoinWorld news, Citrini Research analyst Jukan stated that the improvement in reasoning performance of AI models is not solely achieved by increasing Nvidia GPUs. GPUs often remain idle with low utilization due to memory bottlenecks during reasoning. The ROI of reasoning depends on memory rather than GPUs, and Micron's importance may surpass Nvidia.