−10.6%
since Oct 6
The idea
I believe AI inference is shifting out of the cloud and onto devices people already own, as open models get small enough to run locally. To play that, I own the device makers shipping local AI (Apple is my largest position, betting on an on-device Siri upgrade cycle), the edge chips running those models, the memory and storage that gate performance, and a small sleeve tied to open models. I keep 10% in cash to add on dips, avoid oversized single-name bets outside my top pick, and plan to hold for roughly two years unless the shift toward on-device AI stalls.
What it owns
- Companies selling AI-capable PCs, phones, and consumer GPUs are positioned to benefit first when AI inference moves from cloud servers to hardware people own. · AppleShipping on-device AI features across iPhones and Macs creates a hardware upgrade incentive if those features remain local and roll out consistently rather than staying cloud-dependent., NvidiaConsumer RTX GPUs serve as the standard hardware for running open models locally at home, so demand for these cards depends on local inference remaining a meaningful use case., Dell TechnologiesSelling AI-capable PCs to both business and consumer buyers at scale means Dell's shipment growth will reflect whether local AI features actually drive a hardware refresh cycle., HP Inc.As a major AI PC vendor, HP provides a second way to participate in the device refresh cycle if buyers actually pay for on-device AI capabilities rather than treating them as commodity features., Corsair GamingCorsair sells components for building personal AI rigs, and benefits if running models at home becomes as organized a hobby category as gaming system building., Super Micro ComputerSuper Micro builds small servers that let businesses and serious enthusiasts run models on their own hardware instead of renting cloud capacity, and grows if demand for owned, on-premise AI infrastructure increases.37%
- Local AI only functions if the processor in the device can execute the model efficiently, making chip designers central to whether on-device inference becomes practical. · QualcommSnapdragon chips enabling on-device AI in Android phones and Windows laptops remain in demand if on-device intelligence becomes a genuine purchasing motivation for consumers., Advanced Micro DevicesRyzen AI processors power a significant portion of AI PC shipments, and the take succeeds if the PC refresh cycle materializes and AMD continues gaining share against competitors., Arm HoldingsRoyalties flow to Arm on nearly every edge chip in the basket from Apple to Qualcomm, so Arm benefits if local AI hardware ships in volume on Arm-based designs., BroadcomBroadcom's networking and connectivity chips move data inside and between devices, and higher-capability hardware in homes increases demand for that silicon., Lattice SemiconductorLattice's low-power programmable chips handle AI tasks in small devices that cannot fit a GPU, and this works if inference spreads into cameras, appliances, and sensors as models shrink., AmbarellaProcessing AI vision tasks directly on cameras without cloud connectivity represents the purest form of edge inference, and holds if security cameras, vehicles, and robots keep moving intelligence onto the device.25%
- Running models locally requires substantially more fast RAM and storage per device than prior generations, a demand that device makers often underweight in their component specs. · Micron TechnologyEvery AI-capable phone and PC requires more DRAM than its predecessor to run local models, making memory a steady component of device cost growth., Western DigitalModels and their training data must reside on-device, and storage capacity grows as local AI becomes more capable and prevalent., Seagate TechnologySeagate supplies the bulk storage that holds model files and data that edge AI generates, and benefits from rising total storage demand without requiring heroic growth assumptions., SandiskAs the flash memory pure play, SanDisk benefits when on-device models drive higher storage capacity per phone, laptop, and AI PC.18%
- Shrinking open-source models are what make local AI feasible on consumer hardware, and these companies anchor the supply chain that produces and distributes them. · Meta PlatformsLlama models are the primary driver of open-source AI becoming small and capable enough to run locally, which is the foundation of this entire take., Taiwan SemiconductorManufacturing nearly every chip in this basket from multiple companies means TSMC collects revenue from whichever device maker succeeds, but also concentrates the take's manufacturing and geopolitical risk in one supplier.10%
- Cash10%