Memory Shortage Limits AI Infrastructure; DRAM at 7x P/E Remains Cheap

Across a compressed two-day window last month, the CEOs of Microsoft, Meta, Google, Amazon, and Apple independently cited the same critical bottleneck on their earnings calls: memory capacity is severely constrained and the shortage shows no signs of easing. This coordinated narrative from five of the world's largest AI spenders carries significant weight, as these companies control the infrastructure buildout that will determine AI capability deployments over the next 18 months.

The market's response has been muted despite the severity of the signal. Memory vendors like Micron trade at just 7x trailing earnings, a valuation that appears to discount the persistent demand signal from the largest customers in the sector. Competing narratives about AI capex cycles and peaking spend have clouded investor perception, even as chief architects of AI infrastructure explicitly warn of undersupply.

For semiconductor and infrastructure investors, this divergence creates a tactical opportunity: the CEOs are effectively signaling that memory capex will remain elevated well into 2026 and 2027, supporting margin expansion and pricing power for DRAM and NAND producers. This positioning favours memory-heavy plays like Micron, SK Hynix, and Samsung, while also lifting broader semiconductor capital equipment vendors like ASML and Applied Materials.

The skeptical case rests on cyclicality. History shows that memory capex binges are followed by oversupply and price collapses, and some analysts worry that the current rush by hyperscalers will prove excessive. However, the specificity of the CEO commentary and the scale of AI training infrastructure required to compete globally suggest this cycle may differ from past DRAM downturns.