Storage Stocks Surge 11% as AI Rotation Hits Hardware
Western Digital and SanDisk lead double-digit gains after Gartner upgrades the storage sector to Strong Buy, citing a 40% capacity shortage for AI data lakes.
Data storage stocks exploded Tuesday as institutional investors rotated out of software and into physical AI infrastructure. Western Digital jumped 11%, SanDisk rose 10%, and Seagate gained 7% in what traders are calling the start of a storage supercycle.
The catalyst: a Gartner report upgrading the entire storage sector to Strong Buy, warning that global data storage capacity for AI "data lakes" is currently 40% undersupplied.
The Storage Wall
For two years, AI investment focused on compute—chips, GPUs, and the companies making them. Storage was an afterthought. That's changing.
The "Storage Wall" refers to a growing bottleneck where data processing speeds have outpaced the ability to retrieve and archive information. Training a large language model generates petabytes of intermediate data. Running inference at scale requires rapid access to model weights. Neither works without massive, fast storage infrastructure.
Hyperscalers have noticed. Western Digital reported complete sell-outs of its 40TB UltraSMR drives through fiscal 2026, with Amazon and Google locking up supply for AI training clusters.
The Performance
| Stock | April 1 Gain | YTD Return |
|---|---|---|
| SanDisk (SNDK) | +10% | +181% |
| Western Digital (WDC) | +11% | +89% |
| Seagate (STX) | +7% | +62% |
| Micron (MU) | +4% | +45% |
SanDisk's year-to-date return of 181% reflects its pure-play status after spinning off from Western Digital in February 2025. The company renewed its Kioxia joint venture through 2034, securing flash memory supply for the next decade.
Western Digital, now focused on hard disk drives for nearline and archival storage, saw its shares hit all-time highs. The company has purchase commitments from its seven largest customers extending through 2026, with three extending into 2028.
Why Now
The timing relates to where AI deployment currently sits. Training dominated the first phase—that's compute-intensive. But as companies move trained models into production, the "Inference Inflection" shifts demand toward storage and memory.
Every inference query needs to access model weights. Personalized AI systems require storing user context. Enterprise deployments demand audit trails and version control. All of this is storage-hungry.
Gartner's report quantified what the industry suspected: there simply isn't enough storage capacity being manufactured to meet 2027 demand projections. New fab construction takes 18-24 months. The shortage is structural.
The Broader Rotation
Tuesday's move reflects a broader pivot from software to hardware in AI portfolios. SaaS companies that rallied on AI narratives in 2024 have struggled as investors question their actual AI revenue. Physical infrastructure—chips, storage, power, cooling—offers more defensible moats.
We saw this pattern earlier with Micron's record revenue quarter, driven by HBM demand for AI accelerators. And again when semiconductor stocks hit records on AI spending commitments.
Storage is the third leg of AI infrastructure, and arguably the most undersupplied. Unlike chips, where TSMC and Samsung can add capacity relatively quickly, storage manufacturing involves complex supply chains with long lead times.
What to Watch
Earnings season will test whether storage fundamentals match the stock moves. Micron reports later this month, followed by Seagate in late April. Western Digital and SanDisk report in May.
Pricing power is the key metric. If storage vendors can raise prices without losing volume, margins expand and the rally has legs. If hyperscalers push back, the sector's multiple could compress quickly.
For a deeper look at the sectors driving this rotation, the infrastructure buildout extends beyond storage into power generation, data center cooling, and networking equipment.
The "Software Wall" narrative—that software-only AI plays are hitting limits—may have further to run if hardware constraints persist.