TL;DR
Unconventional AI has released Un-0, an image generation model on a simulated oscillator architecture that founder Naveen Rao says can reduce AI power by a factor of 1,000.
Unconventional AI, a startup founded by former Databricks AI head Naveen Rao, released the first AI modelan imaging system called Un-0 that runs on an entirely new type of computing architecture. According to an accompanying research paper, the model provides results comparable to state-of-the-art diffusion models such as Steady Diffusion. Interestingly, he is working on software simulation of hardware that does not yet exist.
The company is building an oscillator-based computer architecture that forgoes the digital logic that underpins virtually all modern computing. Instead of processing data through transistors performing binary operations, the Unconventional approach uses ring oscillators connected in a fabric network, encoding and processing data through the physics of the oscillators. Rao told TechCrunch that this approach could ultimately reduce power consumption by a thousand times compared to conventional chips.
This claim is voluntary. U.S. utilities plan to spend nearly one and a half trillion dollars on infrastructure by 2030, largely driven by AI data center demand.and any technology that could significantly reduce this burden would be very valuable. But Unconventional did not build a physical chip, and the thousand-fold improvement exists only as a theoretical projection.
What Un-0 demonstrates is that the architecture can replicate the function of conventional artificial intelligence systems. The research team built a fully functional imaging model using software simulation of the oscillator architecture, and the paper shows that it performs on par with established diffusion models. “This is the “hello world” of a new kind of computer” Rao told TechCrunch.
Rao has a track record that makes investors willing to bet on the building. He co-founded Nervana Systems, a deep learning chip startup that Intel acquired in 2016 for about $400 million. He later founded MosaicML, which Databricks acquired in 2023 for approximately one and three billion dollars.
Rao holds a PhD in neuroscience from Brown and studied electrical engineering at Stanford. This background, combining chip design and brain science, is central to his idea that computing architecture itself needs to change.
This record attracted $475 million in seed funding in December 2025, worth four and a half billion dollars, led by Lightspeed and Andreessen Horowitz, with the participation of Sequoia, Lux Capital, DCVC and Jeff Bezos. Rao invested $10 million of his own money on the same terms. It’s not the only startup betting that the path to unconventional AI efficiency lies through fundamentally new architecturesbut his approach is one of the most radical.
The company plans to release schematics of the physical chip soon and intends to build an entire result stack from scratch. The ultimate goal is to act as a computing provider with the ability to supply unconventionals through its own chips. “We will build a new system of our chips,” Rao said, adding that requests will come in and outputs will come out over a standard network connection, but at a fraction of the capacity.
The ambition is too big for the size of the company. Unconventional has fewer than 50 employees and is trying to replace the von Neumann stored-program computer, an architecture that has dominated computing for nearly 80 years. The race to reduce the energy footprint of artificial intelligence has attracted a wave of startupsbut most are working on cooling, efficient software, or incremental hardware improvements rather than trying to completely rebuild the compute stack.
Rao’s argument is that incremental approaches will not suffice. “AI scaling is energy-difficult,“he told TechCrunch, adding that capacity will be the main limit in the next few years. The International Energy Agency predicts that global data center electricity consumption will exceed 1,000 terawatt-hours by the end of 2026.
The gap between a software simulation of Un-0 and a working chip that produces real-world results at scale is vast, and the company has given no timetable for when physical hardware will be available for commercial use. But the demonstration that oscillator-based computing can produce functional AI output is the first concrete evidence that the approach is more than theoretical. Whether it lives up to its promise of a thousandfold efficiency is a question only the hardware can answer.
