Teaching AI to understand the invisible signals that connect every device on Earth.
Superpose builds foundation models for the wireless world. A new sense for machines, alongside language and vision.
§ 1The world we live in
Every space on Earth is full of wireless signals. AI cannot yet read them.
Phones, cars, satellites, drones, factories, hospitals, homes. Every connected device on Earth communicates through electromagnetic signals: radio, Wi-Fi, cellular, radar.
No foundation model today maps wireless data to a structured understanding of an environment: what is present, what is moving, what is anomalous. The information is in the signal. No model has been trained to read it.
Language had its foundation-model moment. Vision had its foundation-model moment. The wireless world has not.
§ 2The bet
Language. Vision. Wireless.
The architectures and training methods behind modern foundation models generalize beyond language and vision. Applying them to wireless data is a question of execution, not invention.
§ 3Why now
The window is open.
- Sensors
- The sensors required to capture wireless signals are now standard hardware in phones, vehicles, and access points. Coverage is global; unit cost has collapsed.
- Technology
- Foundation-model methods are not specific to images and text. The remaining work is data, scale, and engineering.
- Demand
- Telecom, autonomy, aerospace, and national security all require real-time understanding of the physical environment. None of them can get it from cameras alone.
- Field
- No major lab has built one yet. The category is unnamed. New categories produce new leaders.
§ 4What it unlocks
Infrastructure for systems that operate in the physical world.
- Telecom
- Operators diagnose network problems by inference, at the cost of billions per year. A model that reads the wireless environment directly can localize them in real time.
- Autonomy
- Cameras and lidar degrade in fog, rain, and darkness. Wireless signals do not. They carry information about geometry, motion, and material that optical sensors cannot resolve.
- Aerospace
- Aircraft, satellites, and drones operate in dense electromagnetic environments. Most of the available information is currently discarded.
- National security
- Search and rescue, contested airspace, GPS-denied environments: operations like these require sensing that does not depend on optical line of sight.
§ 5The company
Building on a long horizon.
Superpose was founded in 2025 by machine learning and systems engineers from Apple and Meta. The company is pre-seed and building in stealth.