Memory Crunch, AI Supercycle
Desmond Lim@mynameisdesmond
−18.1%
since Oct 7
The idea
I believe AI data centers are soaking up memory and GPU supply, with HBM alone eating about three times the wafer capacity of standard DRAM, and that this squeeze runs for years, not quarters. So this basket leans memory-heavy: the makers who benefit directly from tighter supply and higher prices, the fab tooling and packaging companies that get paid as the industry scrambles to add capacity, and a smaller GPU sleeve since that trade already feels crowded.
What it owns
- New memory factories built to address the shortage must purchase equipment regardless of which memory maker expands capacity. · Lam ResearchIt manufactures the machines that memory factories run on, so new capacity investments by chip makers flow directly to it., Applied MaterialsIt is the broadest equipment supplier to chip fabs, so it collects revenue from nearly every factory built to catch up with current demand., KLA CorporationIt inspects chips for defects, and stacked memory requires more inspection per unit to maintain acceptable yields., ASML HoldingIt is the sole manufacturer of the most advanced chipmaking lithography machines, making it essential to every new fab built during capacity expansions., TeradyneIt supplies test systems that every finished memory chip and GPU must pass through before shipment.35%
- Record prices for memory and storage chips persist while supply remains constrained through the coming years. · Micron TechnologyIt is the only US maker of DRAM and HBM chips, so it captures demand from data centers as memory prices stay elevated while supply remains tight., SanDiskIt is a pure play on NAND flash storage in data center SSDs, which must expand capacity as memory shortages push demand downstream., Seagate TechnologyIt supplies large hard drives to data centers for storing and training on AI workloads, capturing demand as storage requirements surge., Western DigitalIt supplies storage drives to data centers that cannot satisfy the expanding appetite for bulk storage capacity.28%
- Stacking and inspecting high-bandwidth memory chips represents a concentrated bottleneck in the production chain. · Amkor TechnologyIt performs advanced packaging to assemble stacked memory chips into finished products, the step where production capacity is most constrained., Onto InnovationIt inspects stacked memory during manufacturing, and the more difficult the production becomes, the more makers depend on its systems., CamtekIt inspects stacked chips during production, capturing demand as complex memory volumes ramp and manufacturing becomes more difficult., FormFactorIt manufactures probe cards for testing stacked memory, and each new generation of advanced memory requires purpose-built test tooling.16%
- A smaller allocation to the GPU companies whose surging demand is creating the underlying supply squeeze. · Advanced Micro DevicesIt sells AI GPUs competing directly with other makers for the same data center orders pulling capacity away from other applications., Taiwan SemiconductorIt manufactures GPUs for all major designers, so it profits from the growing chip orders regardless of which company wins market share.12%
- Cash9%