Naveen Rao, the former head of AI at Databricks, claims his new company can slash the energy cost of running artificial intelligence by a factor of 1,000.
Unconventional AI, a startup founded by Rao, is betting on a different kind of hardware to achieve this. The firm has released its first model, Un0, which generates images using a software simulation of an oscillator-based architecture.
How it works
The output from Un0 matches the quality of current tools like Stable Diffusion or OpenAI‘s GPT Image 1. The difference lies in the underlying technology. Standard chips used for large language models and image generators rely on a different method of processing data.
Rao’s team built a fully functional image generation model using a simulation of their new chips. The results show performance comparable to state-of-the-art diffusion models. The goal is to replace conventional silicon with this oscillator-based design to drastically cut power consumption.
Next steps
Currently, Un0 runs on a software simulation. The company plans to release schematics for a physical chip soon. The roadmap involves building a complete inference stack from scratch, allowing Unconventional AI to provide compute capacity similar to other cloud providers.
“We will build a new kind of system composed of our chips,” Rao told TechCrunch. “We will run AI models there, and we will have a network cable where prompts come in and inferences go out, but it’ll be done at 1/1000 of power.”
The team consists of fewer than 50 people. Yet, they argue that the massive expansion of AI infrastructure faces a hard limit: available power. As demand for inference grows, energy supply will become the primary constraint.
“AI scaling is hard because of energy. It’s going to be the fundamental limit in the next few years. You just can’t go past it. It’s going to be an energy limited problem, at the end of day,” Rao said.
What it means
For companies running AI workloads, the current trajectory assumes power will remain cheap and plentiful. Rao’s approach suggests that future systems must be designed around severe energy constraints. If the oscillator architecture works as planned, it could allow the same computing power to run on a fraction of the electricity currently required, potentially changing how data centres are built and operated.
