Pateo is betting cockpit AI will need its own hardware, with an add-on box sidestepping the limits of existing vehicle chips
The Pateo AI Box now runs Nvidia’s Nemotron speech recognition and ByteDance’s 35-billion-parameter Doubao language model on Nvidia Drive AGX Thor hardware, keeping core voice functions available without a cloud connection. A leading new energy vehicle maker has nominated the system for a new model due to enter production in the second quarter of next year.
A smaller 2-billion-parameter model sorts and screens requests, and Pateo’s AI Agent platform draws on voice, cameras and sensors to operate cabin controls within set safety permissions. Uses range from spotting driver fatigue on long motorway runs to warning of flooding, fire or parking violations.
Memory matters as much as raw compute. As a trip goes on, the model’s working cache grows with every exchange and task, which Pateo says demands the bandwidth Thor provides.
The box adds dedicated AI computing without replacing a vehicle’s main control chip, giving existing cars whose processors lack headroom for larger models an upgrade route beyond software updates. For automakers, that turns cockpit AI into a hardware add-on that can be fitted across vehicle programmes.
Pateo also wants revenue after the sale. A strategy it calls Tokenomics targets recurring income from AI services, and a June 2026 tie-up with Xunce Technology, Saimo Technology and Mininglamp Technology covers a physical AI world model built around tokens. A nomination from a domestic robotics computing provider takes the platform beyond vehicles.
Source: PR Newswire
Source: www.automotiveworld.com
