Tech news in 3 minutes
Databricks’ former AI chief thinks he can cut AI’s power bill by 1,000x
Unconventional AI, led by former Databricks AI head Naveen Rao, is developing a new oscillator-based computer architecture to make AI inference vastly more power efficient. On Thursday, the company released its first model, Un-0, an image-generation system built on a software simulation of this new architecture. In an accompanying paper, the team shows that Un-0 performs as well as state-of-the-art diffusion models like Stable Diffusion or OpenAI's GPT Image 1. The key advantage is that the oscillator-based computing could reduce power use by up to 1,000 times. The current version runs on a software simulation, but Unconventional plans to release schematics for an actual chip soon, then build an entire inference stack from scratch, eventually offering compute capacity like any other provider. Rao envisions a new kind of system where prompts come in and inferences go out over a network cable, but at 1/1000 of the power. The company has fewer than 50 employees but aims to address what Rao sees as the fundamental limit for AI scaling: energy. He believes that in the next few years, AI will be an energy-limited problem, and Unconventional’s approach is one of the few efforts that can meet that challenge.