−11.6%
since Oct 6
The idea
I believe AI is moving out of the cloud and onto the devices we already own, as open models shrink enough to run fast and private on local hardware. To play that, I built a basket across device makers, edge chips, and memory and storage, with a small sleeve in the open-model ecosystem that publishes the models and fabricates the chips. It is concentrated and meant to be held for a couple of years, not forever, with 10% kept in cash to add on dips. If local inference fails to catch on and everything stays cloud-bound, that's my signal the thesis is wrong.
What it owns
- Companies that sell computers directly to consumers and businesses benefit when AI capability becomes a primary reason for hardware upgrades and replacement cycles. · AppleIt has the largest installed base of devices with dedicated neural processors and has committed to running AI inference on-device rather than in the cloud., NvidiaIts RTX consumer graphics cards are the default hardware for anyone running open AI models on home computers., Dell TechnologiesIt sells corporate laptops and workstations that companies replace when their fleets need on-device AI capability., HP Inc.It is one of the largest PC makers globally, so an upgrade cycle driven by AI capability flows directly through its laptop volumes and pricing., Super Micro ComputerIt builds compact servers that businesses deploy on-premises to run AI locally instead of purchasing cloud capacity., Corsair GamingIt sells memory, power supplies, and cases that enthusiasts buy when building custom PCs to run AI models from home.33%
- Processors designed to run inference locally on phones, laptops, cameras, and edge servers create a new market separate from data center chips. · QualcommIts Snapdragon processors include dedicated NPUs that power AI tasks in phones and Windows-based AI laptops., Advanced Micro DevicesIt ships Ryzen AI processors with built-in neural processing units for laptops, and consumer GPUs for local model inference., BroadcomIt supplies the networking and connectivity chips that edge devices depend on, and custom silicon design for companies building proprietary AI chips., Arm HoldingsIts processor designs power nearly every smartphone and a growing share of laptops, so more AI devices means more royalty revenue., Lattice SemiconductorIt makes low-power programmable chips that handle AI inference in cameras, laptops, and industrial devices where a large processor is unnecessary., AmbarellaIts computer vision chips run AI inference inside cameras and vehicles, with processing happening on the device by design.27%
- Local AI models require more memory and storage capacity per device than prior generations, and this take relies on those spec increases holding as AI devices scale. · Micron TechnologyIt manufactures the DRAM that local AI inference consumes, and AI PCs ship with materially more memory than prior generations., Western DigitalIt manufactures the hard drives that store AI model files and the data they operate on in local setups and home environments., SandiskIt produces the NAND flash memory inside the SSDs that every AI-capable phone and PC requires in larger capacities., Seagate TechnologyIt manufactures drives that store the growing volume of data local AI systems create and consume at home and at the network edge.19%
- Open-source AI models and the manufacturing capacity to support many chip designers are foundational to making on-device AI viable and competitive. · Meta PlatformsIts Llama open models enable capable AI to run on consumer hardware, and releasing competitive open models also serves its own business., Taiwan SemiconductorIt manufactures nearly every chip in this portfolio, from mobile processors to gaming GPUs to AI accelerators, regardless of which designer wins share.11%
- Cash10%