Naveen Rao, the former head of AI at Databricks, has launched a startup called Unconventional AI that he says can reduce the energy consumed by AI inference by a factor of 1,000. The key is a novel oscillator-based computer architecture that replaces traditional transistor-based logic. On Thursday, the company released its first model, Un0, an image-generation system that runs on a software simulation of the new chip design and performs comparably to state-of-the-art diffusion models like Stable Diffusion or OpenAI’s GPT Image 1.
A new kind of computer
Rao told TechCrunch that Un0 represents the “hello world” of a new kind of computing. “This is the ‘hello world’ of a new kind of computer,” he said. “Over the next year, you’re going to start seeing some pretty interesting news around this.” The model’s output is visually similar to existing image-generation tools, but the path to that performance is radically different. Instead of relying on conventional chips that power traditional LLMs, Un0 is built on an oscillator-based architecture that the company believes will ultimately reduce power use by as much as 1,000 times.
Also read: Facebook tests standalone AI companion app to help creators grow audiences
The current version of Un0 runs on a software simulation of Unconventional’s oscillator chips. The company plans to release schematics for an actual chip soon, and from there, build an entire inference stack from the ground up. Rao envisions eventually supplying compute capacity just like any other provider. “We will build a new kind of system composed of our chips,” he 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.”
Why energy efficiency matters for AI
Rao’s focus on power consumption reflects a growing concern in the industry: the available supply of energy may become the hard limit for AI scaling. “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 the day,” he said. As demand for inference—the process of running a trained model to generate output—continues to surge, the cost of powering data centers is becoming a critical bottleneck. Unconventional AI’s approach aims to sidestep that problem entirely by rethinking the underlying hardware.
Also read: Figma adds code layers, animation support, and AI-powered plugin creation in latest update
The company, which still counts fewer than 50 employees, is pursuing an ambitious goal. But given the scale of the AI buildout and the anticipated cost of meeting growing inference demand, Rao believes it may be one of the few efforts capable of addressing the problem at scale.
What’s next for Unconventional AI
The company’s research team published a paper detailing how they built a fully functional image generation model using a software simulation of the new architecture. The results show that the oscillator-based system can replicate conventional AI systems without sacrificing performance. Rao expects to release actual chip schematics soon, followed by a full inference stack that will allow Unconventional AI to offer compute capacity as a service. The long-term plan is to provide a network cable where prompts go in and inferences come out, at a fraction of the power cost of current systems.
Frequently Asked Questions
What is oscillator-based computing?
Oscillator-based computing uses physical oscillators—circuits that produce repetitive signals—to perform calculations, rather than traditional transistors. This approach can drastically reduce energy consumption for certain tasks like AI inference.
When will Unconventional AI’s chips be available?
The company plans to release schematics for an actual chip soon, and eventually build a full inference stack. It aims to offer compute capacity as a service, similar to cloud providers.
How does Un0 compare to other image-generation models?
Un0 produces output similar to models like Stable Diffusion or OpenAI’s GPT Image 1, but uses far less power due to its oscillator-based architecture. It currently runs on a software simulation of the new chip design.
Be the first to comment