Databricks’ former AI chief thinks he can cut AI’s power bill by 1,000x

The race to build the next generation of artificial intelligence has fueled some of the most ambitious projects in tech history. But one startup is betting that the real breakthrough won’t come from a bigger model or more data — it will come from rebuilding the computer itself.

Unconventional AI, led by Naveen Rao — formerly the head of AI at Databricks — is developing a new kind of oscillator-based computer architecture that promises to slash the power required for AI inference by as much as 1,000 times. On Thursday, the company released its first model, called Un0, an image-generation system that demonstrates how its technology can replicate conventional AI systems while using a fraction of the energy.

In an accompanying paper, the company’s research team details how they built a fully functional image generation model using a software simulation of the new architecture. The model performs on par with state-of-the-art diffusion models like Stable Diffusion or OpenAI’s GPT Image 1, but the path to that performance is radically different.

A new kind of computer

“This is the ‘hello world’ of a new kind of computer,” Rao told Bitcoin World. “Over the next year, you’re going to start seeing some pretty interesting news around this.”

The Un0 model runs on a software simulation of Unconventional’s oscillator chips. Unlike traditional CPUs or GPUs that rely on transistor-based logic, oscillator-based computing uses the physical properties of oscillating circuits to perform calculations. This approach can dramatically reduce the energy needed for inference — the process of running a trained AI model to generate outputs.

The company plans to release schematics for an actual chip soon. From there, the goal is to build an entire inference stack from the ground up, eventually supplying compute capacity as a service. “We will build a new kind of system composed of our chips,” says Rao. “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 power matters for AI’s future

The promise of a 1,000x reduction in power consumption is not just a technical achievement — it addresses what many in the industry see as the next major bottleneck for AI scaling. As models grow larger and demand for inference surges, the available supply of power is becoming a hard limit.

“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,” Rao says.

Data centers already consume enormous amounts of electricity, and the growth of AI is expected to accelerate that trend. Major tech companies have announced plans to build new power plants and invest in renewable energy to meet demand. Unconventional AI’s approach could offer an alternative path — one that doesn’t require massive infrastructure buildouts.

From startup to infrastructure provider

Unconventional AI currently employs fewer than 50 people, making its ambition all the more striking. The company is essentially building a new computing paradigm from scratch: new chips, new software, and a new inference stack. The plan is to eventually offer compute capacity just like any other cloud provider, but with dramatically lower energy costs.

If successful, Unconventional AI could reshape the economics of AI inference. For companies running large-scale AI applications, energy costs are a significant and growing expense. A 1,000x reduction would not only lower costs but also enable AI deployment in environments where power is limited, such as edge devices or remote locations.

Conclusion

Unconventional AI’s oscillator-based architecture represents a fundamental rethinking of how computers process AI workloads. While the company is still in its early stages — with a simulated model and plans for a physical chip — its approach addresses a critical challenge facing the industry. As AI continues to scale, the limits of energy supply may prove more binding than the limits of data or model size. If Rao and his team can deliver on their promise, they may have found a way to keep AI growing without breaking the power grid.

FAQs

Q1: What is oscillator-based computing?
Oscillator-based computing uses the physical properties of oscillating circuits to perform calculations, rather than traditional transistor-based logic. This approach can be significantly more energy-efficient for certain types of computations, including AI inference.

Q2: How does Un0 compare to existing image generation models?
According to the company, Un0 performs on par with state-of-the-art diffusion models like Stable Diffusion and OpenAI’s GPT Image 1, but it uses a fraction of the power. The model was built using a software simulation of Unconventional’s oscillator chips.

Q3: When will Unconventional AI release a physical chip?
The company plans to release schematics for an actual chip soon. From there, they aim to build an entire inference stack and eventually offer compute capacity as a service.