The pursuit of unearthing the next breakthrough in AI has funded numerous ambitious initiatives — yet one company views this as an opportunity to reconstruct computing architecture from scratch.

Under the leadership of Naveen Rao, who previously headed AI at Databricks, Unconventional AI aims to enhance inference processing to be significantly more energy-efficient. The unique feature: a novel oscillator-based computer architecture.

On Thursday, the firm unveiled its initial AI model — named Un-0 — a tool for image generation that demonstrates for the first time how the company’s technology can emulate traditional AI systems. In a correlating new publication, the firm’s research team elaborates on how they developed a fully operational image-generation model utilizing a software simulation of the new architecture — one that performs comparably to cutting-edge diffusion models.

“This is the ‘hello world’ of a new type of computer,” Rao stated to TechCrunch. “Over the next year, you’re going to start witnessing some fascinating news around this.”

The production from the fresh Un-0 model resembles that of image-generation frameworks like Stable Diffusion or OpenAI’s GPT Image 1. What stands out is how it achieves that level of performance. The model relies on an oscillator-based architecture that diverges entirely from the chips that drive conventional computing and traditional LLMs. The advantages of the oscillator-based computing are intricate, but Rao is convinced it will ultimately cut energy consumption by as much as 1,000-fold.

Much of the infrastructure required for this goal is still under development. The existing version of Un-0 operates on a software simulation of Unconventional’s oscillator chips, but the company intends to share designs for a tangible chip soon. From that point, the strategy is to construct a complete inference stack from the ground up, with Unconventional AI eventually providing computing capacity just like any other supplier.

“We will create a new type of system made up of our chips,” says Rao. “We will execute AI models there, and we will have a network cable where prompts are inputted and inferences outputted, but it’ll be accomplished at 1/1000 of the power.”

It’s an extraordinarily ambitious objective, especially for a company with fewer than 50 employees. However, considering the magnitude of the AI expansion and the expected costs of satisfying the rising demand for inference, it may represent one of the few initiatives capable of addressing the scale of the challenge. As Rao perceives it, the available energy supply will be one of the crucial limits for AI in the coming years — and Unconventional is among the rare projects that can tackle it.

“AI scaling is challenging due to energy. It will be the fundamental constraint in the next several years. You simply can’t surpass it. Ultimately, it’s going to be an energy-constrained issue,” he remarks.