The 1,000× Power Play — Ex-Databricks AI Chief Rebuilds Computing From Scratch

Naveen Rao's Unconventional AI just shipped its first model — an image generator running on oscillator chips that could slash inference energy by three orders of magnitude. No physical silicon yet, but $475M in backing says the bet is real.

The most radical bet in AI hardware just got its first public proof of life — and it doesn't use transistors the way the rest of the world does.

Naveen Rao, the former head of AI at Databricks, has shipped Un-0, the first functional model from his startup Unconventional AI. It generates images. It performs on par with state-of-the-art diffusion models. And it runs on an architecture that abandons the digital logic gate — the foundational building block of every computer for the last 80 years.

Instead of processing data through transistors performing binary operations, Unconventional's approach uses coupled ring oscillators in a fabric network. Information is encoded and processed through the physics of the oscillators themselves — not through software running on general-purpose silicon.

"This is the 'hello world' of a new kind of computer," Rao told TechCrunch.

The Numbers That Make This Matter

The pitch isn't subtle. Rao claims the oscillator architecture will ultimately reduce AI inference power consumption by as much as 1,000 times compared to conventional chips.

The claim is aspirational — Un-0 currently runs on a software simulation of the oscillator architecture, not on physical silicon. The company has yet to fabricate a chip. But it's also not coming from nowhere.

Rao has the kind of track record that makes investors write very large checks: he co-founded Nervana Systems (acquired by Intel for approximately $400 million in 2016) and MosaicML (acquired by Databricks for approximately $1.3 billion in 2023). He holds a PhD in neuroscience from Brown and studied electrical engineering at Stanford — a background that bridges chip design and brain science.

That resume attracted $475 million in seed funding at a $4.5 billion valuation in December 2025, led by Lightspeed and Andreessen Horowitz, with participation from Sequoia, Lux Capital, DCVC, and Jeff Bezos. Rao invested $10 million of his own money at the same terms.

Why Now

"AI scaling is hard because of energy. It's going to be the fundamental limit in the next few years," Rao said. "You just can't go past it."

The numbers back him up. The International Energy Agency projects global data center electricity consumption will exceed 1,000 terawatt-hours by the end of 2026. US utilities alone are planning nearly $1.5 trillion in infrastructure spending by 2030, driven largely by AI data center demand.

Unconventional AI — still under 50 employees — plans to release chip schematics soon and build an entire inference stack from the ground up. The end goal: operate as a compute provider where "prompts come in and inferences go out, but it'll be done at 1/1,000 of power."

Most competitors are chasing incremental efficiency — better cooling, smarter workload scheduling, marginally faster chips. Rao is trying to replace the von Neumann architecture itself.

The Catch

There is no working chip. The 1,000× claim exists only as a theoretical projection. The gap between a software simulation generating images and a commercially viable inference chip running trillion-parameter models at scale is vast, and no timeline has been given.

But for the first time, there's concrete evidence the approach is more than a white paper. Un-0 proves oscillator-based computing can produce functional AI output. Whether it can deliver on the power promise — and whether it can do so before the energy crisis Rao is betting on arrives — is the question that $475 million is now racing to answer.

Sources: TechCrunch, TNW, The AI Insider