Big_Timer
Oct 4, 10:44 PM
$MU 🔥 AI DATACENTERS DON’T RUN ON GPUs ALONE — THEY RUN ON MEMORY + STORAGE.
My top 5 AI memory/storage plays:
MU — The purest U.S. AI memory play. HBM + DRAM demand keeps exploding as AI servers require more bandwidth and capacity.
$SNDK — NAND is becoming a critical layer for AI inference, KV cache and massive datasets. Flash demand should keep rising as AI workloads scale.
$WDC — AI generates mountains of data that have to live somewhere. Training datasets, inference logs, embeddings and outputs create recurring storage demand long after the GPU is installed.
$STX — HAMR + hyperscale storage demand could be a monster combination. AI datacenters are driving enormous long-term storage requirements.
$SIMO — The sleeper. Enterprise SSD controllers are becoming increasingly important for AI infrastructure, KV-cache offload and near-GPU storage.
Everybody chases the GPU.
I’m watching the companies feeding those GPUs with HBM, DRAM, NAND and massive amounts of storage.
AI compute keeps scaling → memory per server rises → inference explodes → stored data compounds.
That’s a MULTI-YEAR infrastructure cycle.
2 replies
