Naveen Rao Wants to Slash AI Power Consumption by 1,000x

The company’s debut image-generation tool, Un-0, mirrors the performance of industry-standard diffusion models like Stable Diffusion or OpenAI’s image generators. While currently running on a software simulation, the results serve as a proof-of-concept for a hardware stack built from the ground up. Rao envisions a future where Unconventional AI supplies compute capacity through proprietary chips, processing prompts with a fraction of the electricity required by existing data centers.

Energy scarcity remains the primary bottleneck for the scaling of large language models. With a team of fewer than 50 people, the startup aims to address this physical constraint by replacing the chips that power traditional computing with oscillator-based systems. Rao maintains that as power supply becomes the definitive hard limit for technological growth, shifting the fundamental architecture of inference processing is the only viable path to maintain current AI development trajectories.