I've been building a small position in AI storage stocks for 3 months, here's why I think storage is the most overlooked layer of the AI

Everyone talks about Nvidia. Nobody talks about where all that AI-generated data actually lives.

Three months ago I started building a small basket, just enough skin in the game to stay sharp, around what I call the storage layer of the AI stack. Here's how I think about it.

**The thesis in one line**

Every AI model trains on data, outputs data, and stores data. Compute gets all the attention. Storage is the quiet infrastructure underneath it all, and it's nowhere near priced in.

**How I split the basket**

**MU (Micron) 30%** The only US-based pure-play memory manufacturer doing both DRAM and NAND. The HBM angle is real, every Nvidia H100 and B100 has High Bandwidth Memory inside it, and Micron is ramping hard. Most people know Micron as a cyclical memory trade. I think the AI supercycle is structurally different because demand isn't coming from one sector, it's coming from every hyperscaler simultaneously.

**SNDK (SanDisk) 25%** Just spun off from Western Digital as a pure-play NAND flash company. The spinoff is the thesis, WDC was always a mixed bag of HDD (dying) and NAND (growing). Now SNDK can allocate capital purely toward flash. Enterprise SSDs for AI data pipelines are the growth driver. It's early post-spinoff so price discovery is still happening.

**LRCX (Lam Research) 25%** Not a storage company per se, they make the equipment that manufactures memory chips. If MU and SNDK are building new fabs, Lam sells the shovels. High margin, recurring equipment spend, and a direct leverage play on memory capex without the inventory cycle risk.

**PSTG (Pure Storage) 20%** The enterprise storage infrastructure layer. Their FlashArray products are going into AI data centers at scale. Unlike the others, Pure has a recurring subscription revenue model which gives it more predictable cash flows. It's the least cyclical of the four.

**Three months in, honest take**

MU has been the most volatile, as expected. SNDK is still finding its footing post-spinoff. LRCX has been the steadiest. PSTG has quietly outperformed.

The position is small, this was never about making money fast, more about building conviction through having real exposure. I track the thesis quarterly: is AI data storage demand growing? Are these companies winning their respective layers? So far both answers are yes.

Curious if anyone else is thinking about AI from the storage angle rather than just the compute angle. Feels underdiscussed relative to how critical it is.

PS: I have small positions in all of the above mentioned