A startup run by one of Databricks’ former AI leaders just showed off its first model, and the pitch behind it is hard to ignore: run AI inference at one-thousandth of the power it takes today. According to TechCrunch AI, the company is Unconventional AI, led by Naveen Rao, who previously headed AI at Databricks. The bet isn’t a faster chip or a leaner model. It’s a rebuild of computing from the ground up.
On Thursday, Unconventional AI released its first model, called Un-0. It’s an image-generation system, and Rao describes it as proof that the company’s approach can match the AI tools we already use. TechCrunch AI reports the team also published a paper detailing how they built a fully functional image generator on a software simulation of a brand-new architecture, one that performs on par with state-of-the-art diffusion models.
What stands out here isn’t the pictures. It’s how the model gets to them.
What Unconventional AI actually launched
- Un-0, an image-generation model. Its output looks similar to Stable Diffusion or OpenAI’s GPT Image 1. The point of the release is the method underneath, not the visuals.
- An oscillator-based architecture. This is a different kind of computer entirely, not the chips that run conventional computing or today’s large language models. Rao believes it can eventually cut power use by as much as 1,000 times.
- A research paper. The team showed the model running on a software simulation of their oscillator chips, matching the quality of leading diffusion systems.
- A roadmap to real hardware. The company plans to release schematics for an actual chip soon, then build a full inference stack on top of it.
Rao called Un-0 “the ‘hello world’ of a new kind of computer” and told TechCrunch AI, “Over the next year, you’re going to start seeing some pretty interesting news around this.”
How it’s meant to work in practice
The long-term plan is for Unconventional AI to become a compute provider like any other. Prompts go in, inferences come out, just at a fraction of the energy cost. “We will build a new kind of system composed of our chips,” Rao said. “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.”
That’s the vision. The reality is earlier-stage. The current version of Un-0 runs on a software simulation, not physical oscillator chips. Much of the infrastructure to reach the 1,000x claim still has to be built.
The caveats worth keeping in mind
- No physical chip yet. Un-0 runs on a simulation. The schematics for real silicon are still coming.
- A small team chasing a huge goal. The company counts fewer than 50 employees. Rebuilding computing architecture is not a small lift.
- The 1,000x figure is a projection. It’s what Rao believes the architecture can ultimately deliver, not a benchmarked result on shipping hardware.
Why this matters
This is significant because of the problem it targets. The AI buildout is colliding with the limits of available power, and inference demand keeps climbing. Rao frames energy as the hard ceiling on AI’s growth. “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,” he told TechCrunch AI.
Most efforts to ease that crunch work within the existing chip paradigm: better GPUs, smaller models, smarter scheduling. Unconventional AI is doing something rarer. It’s trying to change the machine itself. If oscillator-based computing delivers even a fraction of the promised efficiency, it reshapes the economics of running AI at scale. If it doesn’t, it joins a long list of ambitious architectures that never left the lab.
Either way, Un-0 is the first real signal. The next year, by Rao’s own framing, is where the harder proof has to show up. For the full technical details and the team’s paper, the original report at TechCrunch AI is the place to look.