Yes, the data movement between compute and memory is actually the constraint. The CXL-based memory pool built on superior connectivity will boost utilisation of GPUs/AI accelerators and has to be implemented as the next step. Nice write up as always. Super bullish on $MRVL
