AI Moves Into Your Pocket
Wyatt Miller@wyatt
−9.8%
since Oct 2
The idea
I believe frontier AI models are about to get small and efficient enough to run directly on phones and laptops within the next year or two, instead of staying locked in data centers. To play that, I own the chips doing on-device inference and the memory they depend on as the core of the basket, with a smaller slice in device makers who benefit once the upgrade cycle follows, plus foundry and design exposure underneath it all.
What it owns
- These companies make the processors that run AI models directly on phones and laptops, so they capture revenue when models move from data centers to devices. · QualcommIts Snapdragon chips are the primary way advanced AI runs on Android phones, capturing licensing fees when on-device inference becomes standard in flagship devices., Arm HoldingsIt collects a royalty on nearly every phone and growing share of laptops, so a surge in AI-capable devices pays it regardless of which chipmaker wins the on-device race., NVIDIAIts chips power AI on PCs and edge devices running models locally, expanding its revenue beyond data centers if frontier models shrink onto personal hardware., AmbarellaIt builds low-power vision AI chips that run directly on cameras and small devices, offering a pure play on efficient models spreading into edge hardware., NXP SemiconductorsIt puts small AI chips in cars and industrial equipment that cannot rely on cloud connections, capturing revenue if on-device inference spreads beyond phones into embedded systems., SynapticsIt supplies connectivity and processing chips inside phones, earbuds, and smart devices, benefiting when AI features on those devices drive an upgrade cycle., Lattice SemiconductorIt makes ultra-low-power chips that let AI run on devices with minimal battery drain, positioning it where demand appears if models become efficient enough for the smallest hardware.39%
- Running large models on devices requires fast, low-power memory that current hardware doesn't supply enough of, creating a less obvious but critical constraint in this shift. · Micron TechnologyRunning large models locally requires far more fast, low-power memory than phones currently carry, and it supplies exactly that type of memory to device makers., RambusIt licenses technology that moves data between chips and memory faster, solving the exact bottleneck that emerges when devices run large models locally., SanDiskModels stored on devices must live in on-device storage, and larger models mean bigger storage per device, which it supplies to phone and laptop makers.20%
- If on-device AI becomes the reason customers replace their phones and laptops sooner, these device makers capture a wave of upgrades across their installed base. · AppleIt is committed to running AI on-device rather than in the cloud and has a customer base large enough to turn that preference into a replacement cycle., Dell TechnologiesIt sells AI-capable laptops through its corporate channel where business upgrades would materialize if local models become a buying driver for refreshes., HPIt ships huge volumes of consumer and business PCs, so even a modest uptick in AI-driven upgrades moves its numbers significantly over time.16%
- Every new on-device AI chip gets manufactured or designed with these companies' tools and services, so they benefit regardless of which chipmaker leads the on-device race. · Taiwan SemiconductorIt manufactures most of the chips in this take, from Qualcomm to Apple to NVIDIA, so it gains from whichever on-device AI chipmaker captures the market., Cadence Design SystemsEvery new on-device AI chip gets designed with its tools, so a wave of edge AI silicon means sustained demand for its design software and services.15%
- Cash10%