Nvidia's $279 billion memory warning, and why Micron wins

The article frames the current memory market as a squeeze with two engines. Training runs are getting bigger and inference is moving into production, so each new rack of accelerators needs more bandwidth and more bits sitting closer to the GPU. On the supply side, wafer capacity is limited and converting factory lines to high bandwidth memory is presented as tedious and costly, while NAND supply is tight for a different reason linked to hyperscale storage requirements.

That context sets up the article's comparison. Nvidia, having flagged memory as a strategic constraint rather than a background cost, put a multi-year claim on memory bits, and the article argues this spend is aimed at stacked memory and DRAM rather than flash. Micron sits on both of those layers, while Sandisk rides the same AI data-center wave from the storage side. The piece also notes the valuation gap: Micron's forward price-to-earnings ratio hovers around 6, which the article attributes to the market treating memory as a boom-bust commodity, and argues that sustained AI infrastructure demand could let earnings stay high long enough to change that view.

The stakes appear to hinge on whether the demand Nvidia describes stays unconstrained and whether hyperscaler capex keeps treating memory as a first-order constraint. If it does, the article's case for Micron as the closer-to-the-bottleneck pick is likely to hold up; if memory pricing cools or supply catches up, the cyclical discount the market applies to the category may prove harder to shake.