Memory Stocks Up 211-653% YTD: Is the Boom-Bust Cycle Finally Dead?

Here is a stat that will make you do a double-take: $SNDK (SanDisk) is up over 600% year-to-date. $MU (Micron) and $WDC (Western Digital) aren't far behind, posting extraordinary gains in a range of 211% to 653% across the memory and data storage group. The question every rational person has to ask is: is this AI euphoria, or is something genuinely structural happening?

The bull case, made loudly by Jim Cramer on CNBC this week, is that AI data center demand has permanently broken the old boom-bust cycle that defined memory for decades. Memory makers have reportedly adopted disciplined supply practices, locking in long-term customer agreements and posting record gross margins at both SanDisk and Micron. The logic is simple: hyperscalers need more storage than they can get, and they're willing to sign long-term deals to secure it. That's a different world from the spot-market-driven chaos of prior cycles.

The catalyst lighting a fire under the sector this week was Anthropic's blowout revenue numbers. The AI company reported preliminary Q2 revenue with an annualized run rate crossing $6.5 billion, a roughly seven-fold increase year-over-year. When an AI lab is scaling that fast, it needs compute, and compute needs memory. SanDisk popped nearly 9% in a single session on the news. $MRVL (Marvell) added 5.5% and $WDC gained 5.3% on the same day, with the whole sector riding the AI infrastructure narrative.

The bear case is quieter but worth hearing. Memory stocks have been called structurally different before, usually right before the cycle turns. The 30-year Treasury yield just hit its highest level in roughly 25 years, and there are whispers out of Japan about a potential $620 billion selloff of US equities to defend the yen. Elevated borrowing costs and a macro shock would pressure even the most disciplined memory suppliers if enterprise customers suddenly tighten budgets.

What to watch: Walmart and Target report later this week as key consumer health checks. If the consumer is cracking, discretionary tech spending could follow, eventually hitting the hyperscalers' capex confidence. But for now, the AI infrastructure buildout, described by some analysts as adding 20 gigawatts of computing capacity annually at roughly $50 billion per gigawatt, shows no sign of stopping.

The old playbook said memory was a commodity business with commodity timing. The new playbook says it's critical infrastructure. One of those playbooks is about to be wrong, and the market has placed its bet.