−9.6%
since Oct 6
The idea
I believe AI inference is moving out of the cloud and onto the devices people already own, as open models get small enough to run locally for privacy, speed, and cost. So this owns the device makers, the chipmakers designing edge silicon, and the memory and storage that local models consume. On top of that I'm betting personal agents will live at the edge too, so whoever manages the most of them owns the moments of intent, which is why platforms like Meta, Google, Microsoft, and Amazon get a real slice.
What it owns
- These companies sell the phones, laptops, and home hardware that people upgrade to get on-device AI capabilities. · AppleIt sells the hardware people already own and runs models locally for privacy and speed across iPhone and Mac., NvidiaIts chips power AI PCs and edge boxes, earning from inference that moves out of the data center to devices., Dell TechnologiesIt sells the AI PC refresh to businesses that need new machines with AI silicon to run local inference., HP Inc.It offers cheap, broad distribution of AI-capable laptops to consumers who replace older machines for new capabilities., Corsair GamingIt sells to enthusiasts who run local models at home first, people buying RAM and cooling to do it themselves., Super Micro ComputerIt builds the edge and on-prem servers for inference that leaves the big clouds but doesn't fit on a laptop.29%
- These are the chips that actually run local inference, plus the foundry that manufactures nearly all of them. · QualcommIt designs the chips that run AI on phones and laptops, which is exactly where inference is heading., Advanced Micro DevicesIt puts AI accelerators inside laptop processors, earning a cut on every AI-capable laptop regardless of the brand., Taiwan SemiconductorNearly every edge AI chip gets made at its fabs, so it collects from the growth in edge silicon demand across all platforms., Arm HoldingsIts designs sit under almost every phone and a growing share of AI PCs, collecting a royalty on the whole shift., BroadcomIt supplies the connectivity chips inside devices and builds custom AI silicon for major platforms across the ecosystem., Lattice SemiconductorIt makes tiny low-power chips for AI at the far edge, in cameras and sensors and always-on features., AmbarellaIts chips do vision AI on the device itself with no cloud round trip, embodying the privacy and speed argument in hardware.24%
- Whoever runs a personal agent owns the moments of intent and the surrounding context, which is where the monetization happens. · Meta PlatformsIt gives Llama away so agents run on its models everywhere, then monetizes the moments of intent through its ad platform., AlphabetIt puts Gemini on billions of Android phones, which is the distribution needed to own moments of intent at scale., MicrosoftIt owns the work context through Copilot inside documents, email, and meetings, converting that presence into paid seats., AmazonIt already owns the moment of buying intent, and Alexa plus its devices position agents at the edge of shopping.21%
- Running large language models locally requires substantial fast memory and quick storage, so devices performing edge compute need increasing amounts of both. · Micron TechnologyIt sells the RAM that local models require, and benefits if devices start shipping with more memory to run AI locally., Western DigitalIt sells the storage that holds model weights and the data they work on at the edge of the network., SandiskIt is the pure flash play, providing the fast storage inside phones and laptops that local models load from., Seagate TechnologyIt is the cheaper, bulkier storage backstop for edge servers and home setups that keep local data local for privacy.16%
- Cash10%