Material Intelligence Suppliers
Clay Norris@claynorris
−6.2%
since Oct 1
The idea
I believe AI's real unlock is turning raw physical signal, photons, vibration, echoes, into usable data, which raises the value of everything that captures it. Instead of betting on one sensor type or model winner, I own suppliers across the whole stack: image sensors, MEMS and signal chain, medical and ultrasound imaging, premium sensing like lidar, and the edge silicon that runs inference near the sensor. Large caps anchor it, smaller pure plays carry the upside, and I keep some cash on the side.
What it owns
- Cameras are the highest-volume sensors on earth, and AI vision models make every pixel they capture more valuable. · Sony GroupIt makes more of the world's image sensors than anyone else, so if AI keeps raising the value of every captured photon, Sony sells the capture., onsemiIt supplies the image sensors cars use to see, the pure-vision side of the bet that software now extracts what lidar-versus-vision arguments used to fight over., CognexIt sells complete machine vision to factories, pairing cameras with software that reads them at the moment modern models make raw factory imagery usable., Teledyne TechnologiesIt makes specialty sensors in infrared, x-ray, and scientific imaging that widen the kinds of physical signal AI can learn from beyond visible light.23%
- Tiny motion, sound, and pressure sensors plus the chips that digitize their signals are the plumbing of every material intelligence product. · Analog DevicesIt sits between the physical world and the model, converting vibration, voltage, and temperature into digital signal that AI can use., TE ConnectivityIt puts sensors and the connectors that carry their signal into cars, factories, and medical devices, bridging the data flywheel across industries., STMicroelectronicsIt ships the high-volume MEMS motion sensors that flood the world with the entropy software can now interpret as accelerometers and gyros., KnowlesIt makes the MEMS microphones that capture audio and vibration signal models can now decode across hearing, industrial, and medical uses.21%
- Someone has to run the model next to the sensor, and these chips turn raw signal into answers on the device itself. · QualcommIt runs models next to the sensor in phones, cars, and devices, turning captured entropy into answers without a round trip to the cloud., AmbarellaIt builds chips that run vision AI right at the camera, the purest edge-inference bet where inference happens closest to the sensor., Lattice SemiconductorIt makes tiny, low-power programmable chips that sit beside sensors and pre-process their signal as volumes of sensing hardware grow., NXP SemiconductorsIt embeds processors in cars and industrial gear where sensor data actually lives, anchoring the edge silicon layer in production vehicles and machines.20%
- The high-channel-capacity camp: lasers, photonics, and lidar that capture cleaner signal and win if raw fidelity still matters. · Hesai GroupIt ships lidar at real volume in the premium channel-capacity camp, participating in the bet regardless of which sensing approach wins in production., CoherentIt supplies the lasers and photonics under lidar, optical sensing, and the data links that move captured signal across systems., NovantaIt makes precision photonics and motion components inside medical and industrial sensing gear, a picks-and-shovels layer as equipment makers design more sensing in., OusterIt builds digital lidar on standard chip processes, betting that premium sensing can ride semiconductor cost curves down in robotics and industrial uses.15%
- Ultrasound and body-signal hardware is where AI interpretation is reopening problems the previous decade gave up on as too noisy. · GE HealthCareIt is the incumbent in ultrasound, extracting information from noisy echo exactly as AI now makes that signal interpretable., Koninklijke PhilipsIt pairs ultrasound with patient monitoring, owning two streams of noisy physiological signal that better software makes more valuable., Butterfly NetworkIt put ultrasound on a semiconductor chip, betting the sensor can be cheap and noisy because software does the interpreting.14%
- Cash8%