Every $1 $NVDA makes → $0.50 flows into these 10 stocks:
$TSM — $0.12-0.15 per dollar. All chips fab here. Advanced nodes + CoWoS packaging = bottleneck. No TSMC, no GPUs. Tightest lock, most stable cash flow.
$SMCI — $0.08-0.12 per dollar. NVDA sells chips, customers buy racks. Supermicro does assembly, testing, delivery. GPU demand = rack orders.
$MU — $0.05-0.08 per dollar. HBM memory now costs more than the GPU substrate. Only 3 suppliers globally. More GPUs shipped = more memory bills.
$AVGO — $0.04-0.06 per dollar. Networking silicon. You need switches to turn thousands of GPUs into one supercomputer. Customers pay separately, but without this layer, cards can't cluster.
$VRT — $0.03-0.04 per dollar. Power + liquid cooling. Expensive GPUs = power-hungry + heat-sensitive. No power infra = dead metal. More NVDA shipments = more cooling installs.
$SNDK — $0.02-0.03 per dollar. Storage scales with cluster size. Not as big as TSMC or Micron per unit, but rides volume.
$MRVL — $0.02-0.03 per dollar. Once their interconnect design gets into the rack, every new unit = repeat purchase. NVDA growth compounds down through Marvell's chips.
$VST — $0.02-0.04 per dollar long-term. Electricity layer. Not on NVDA's purchase order, but shows up in power bills + grid capex. More cards shipped = tighter power = generator margin expansion.
$CRDO — $0.01-0.02 per dollar. High-speed connectivity. Strongest GPU is useless if isolated. Every new rack row = another round of connector sales. Small per unit, tight to shipment volume.
$AAOI — $0.01-0.02 per dollar. Optical modules (800G). GPUs move data via optics. More cards = more optical ports needed. Eats the internal data transfer tax.
Copy NVDA's supply chain playbook.
$TSM — $0.12-0.15 per dollar. All chips fab here. Advanced nodes + CoWoS packaging = bottleneck. No TSMC, no GPUs. Tightest lock, most stable cash flow.
$SMCI — $0.08-0.12 per dollar. NVDA sells chips, customers buy racks. Supermicro does assembly, testing, delivery. GPU demand = rack orders.
$MU — $0.05-0.08 per dollar. HBM memory now costs more than the GPU substrate. Only 3 suppliers globally. More GPUs shipped = more memory bills.
$AVGO — $0.04-0.06 per dollar. Networking silicon. You need switches to turn thousands of GPUs into one supercomputer. Customers pay separately, but without this layer, cards can't cluster.
$VRT — $0.03-0.04 per dollar. Power + liquid cooling. Expensive GPUs = power-hungry + heat-sensitive. No power infra = dead metal. More NVDA shipments = more cooling installs.
$SNDK — $0.02-0.03 per dollar. Storage scales with cluster size. Not as big as TSMC or Micron per unit, but rides volume.
$MRVL — $0.02-0.03 per dollar. Once their interconnect design gets into the rack, every new unit = repeat purchase. NVDA growth compounds down through Marvell's chips.
$VST — $0.02-0.04 per dollar long-term. Electricity layer. Not on NVDA's purchase order, but shows up in power bills + grid capex. More cards shipped = tighter power = generator margin expansion.
$CRDO — $0.01-0.02 per dollar. High-speed connectivity. Strongest GPU is useless if isolated. Every new rack row = another round of connector sales. Small per unit, tight to shipment volume.
$AAOI — $0.01-0.02 per dollar. Optical modules (800G). GPUs move data via optics. More cards = more optical ports needed. Eats the internal data transfer tax.
Copy NVDA's supply chain playbook.