Memory Shortage Constrains AI Capex Despite Cheap Valuations: MU, NVDA

The narrative crystallized in a two-day window last month when executives from Microsoft, Meta, Alphabet, Amazon, and Apple each independently flagged the same problem: memory availability is the bottleneck throttling their AI infrastructure expansion. The consistency of messaging across five of the world's largest tech companies signals a material supply constraint, not isolated commentary.

Micron Technologies remains the primary beneficiary of this bottleneck, yet the market prices it at just 7x trailing earnings, a significant discount to both historical averages and the growth profile implied by near-insatiable hyperscaler demand. This disconnect suggests two competing theses: either memory constraints are temporary and will ease as supply catches up (bearish for valuations), or the market fears AI capex peaks soon and memory demand reverts to normal historical patterns.

The implication ripples across semiconductor and infrastructure plays. NVIDIA maintains its position as the foundational compute layer, but memory module makers like Micron, SK Hynix, and Samsung Electronics face pressure to prove demand sustainability beyond the current cycle. Similarly, data center operators and AI infrastructure REITs benefit from confirmed capex momentumThe empirical fact that winners keep winning over the medium term., but the memory bottleneck could force a lengthening of deployment timelines and higher marginal costs for builders.

Skeptics argue that 7x earnings on Micron reflects appropriate caution: memory historically oversupplies and undergoes violent pricing deflation. Previous cycles show that once supply comes online, margins compress rapidly. The fact that the stock refuses to re-rate despite CEO validation of demand suggests the market has priced in a probability of mean reversion that many bulls dismiss.