Former Databricks AI chief Naveen Rao is backing a bold new direction in computing through his startup, Unconventional AI. The company is aiming to make AI inference dramatically more efficient by rethinking the hardware layer from the ground up.
Its first public model, Un-0, is an image-generation system designed to demonstrate how an oscillator-based architecture can match the performance of modern diffusion models while using a very different computing approach. The team says the model was built through a software simulation of its chip concept, offering a first look at the platform's potential.
Rao describes the project as the starting point for a new kind of computer architecture. Instead of relying on conventional chips, Unconventional AI is developing oscillator-based systems that it believes could eventually reduce power consumption by up to 1,000 times.
The company is still early in its journey, with fewer than 50 employees, but it plans to publish chip schematics and build a full inference stack around its own hardware. The long-term vision is to deliver AI compute as a service, with prompts entering and results coming out through a dedicated system built for efficiency.
As AI demand grows, energy use is becoming a central design challenge. If this architecture scales as intended, it could help shape a more sustainable foundation for future AI infrastructure.