On-Device AI Supply Chain
Chris Muir@cmuir
−12.6%
since Sep 30
The idea
I believe AI is moving out of the cloud and onto the devices we already carry, and that shift rewards the physical supply chain behind it more than any single app or model. So this basket spreads across the layers: chip designers, memory makers (since local AI is bottlenecked by RAM and bandwidth, not just compute), the foundries and packagers that fabricate everything, and the equipment makers every fab needs. I'm holding across the whole chain rather than picking one winner, keeping about 10% in cash to add on dips, with a 2-3 year horizon for the shift to play out.
What it owns
- Semiconductor fabs and memory production lines require specialized equipment and materials from these suppliers before they can begin operations. · ASML HoldingIt is the sole producer of extreme ultraviolet lithography machines needed to print the most advanced AI chips, making it indispensable to any fab expansion., Applied MaterialsIts broad equipment portfolio captures orders from fabs expanding capacity to produce more AI chips and memory for on-device workloads., Lam ResearchIts tools are heavily used in memory manufacturing, so demand for its equipment rises as device makers require more DRAM and storage per phone or laptop., KLA CorporationAs chips advance to support on-device AI, catching defects in smaller and more complex designs becomes harder and more valuable, strengthening its inspection tools business., EntegrisIt supplies the ultra-pure materials and handling systems that every fab consumes daily, earning based on overall silicon production volume rather than any single design winning.23%
- These companies design the chips that execute AI models directly on phones, laptops, and edge devices rather than sending data to the cloud. · QualcommIts Snapdragon processors with integrated neural engines are the primary vehicle for running AI models locally on Android phones., Arm HoldingsIts architecture powers nearly every smartphone and a growing share of laptops, so licensing demand increases as chip designers add AI capabilities to their Arm-based designs., Advanced Micro DevicesIts Ryzen AI chips enable models to run locally on laptops, positioning it on the processor design side of on-device AI adoption in the computing market.20%
- These companies fabricate and assemble the AI chips that processor designers create, handling the physical production and integration into devices. · Taiwan SemiconductorIt manufactures nearly every advanced processor in this supply chain, so it captures value across the entire ecosystem as on-device AI production scales., GlobalFoundriesIt manufactures the support chips found in phones and edge devices like radios and sensors, which multiply in volume when AI moves onto the device., Amkor TechnologyAdvanced packaging work increases as AI chips and memory densify inside phones and laptops, and it is one of the largest independent assembly providers., ASE TechnologyIt assembles and tests a large portion of global chip production, earning revenue on every AI component that ships regardless of which designer wins.20%
- Running models locally requires far more RAM and storage per device, so manufacturers need to install larger memory chips in phones and laptops. · Micron TechnologyAs one of only three global DRAM makers, it supplies the memory that becomes the limiting factor for local models, with demand rising as devices pack more RAM., Western DigitalIt supplies storage solutions used when more data creation and retention occurs locally on devices, benefiting from rising total storage demand in the ecosystem., SanDiskAI models living on the device must be stored on the device, and its flash memory is where they reside in phones and laptops.16%
- These smaller chipmakers focus on low-power AI processing for cameras, cars, and gadgets beyond phones, enabling intelligence across the broader device ecosystem. · NXP SemiconductorsIt places AI-capable chips in cars and industrial equipment, capturing the portion of the shift where edge devices gain local intelligence rather than relying on cloud processing., Lattice SemiconductorIts tiny, ultra-low-power chips let battery-constrained devices run AI inference locally, addressing the core power constraint that on-device models face., AmbarellaIts camera and video processing chips enable devices to analyze footage locally instead of in the cloud, representing one of the clearest use cases for on-device AI., Ceva IncIt licenses neural processing architectures that chipmakers integrate into their designs, earning royalties on many devices without manufacturing anything itself.12%
- Cash10%