This AI cycle is fundamentally different from every prior memory supercycle in $MU history. Past supercycles were driven by unit volume growth with more phones + servers buying largely commoditized DRAM but but this one is a capacity-constrained cycle where HBM sells at ~5x conventional DRAM ASPs. Hyperscalers are willing to pay whatever it takes because the real cost is leaving massive GPU clusters underutilized. That is how Micron ends up producing $16B of operating profit in a single quarter.